Search

In this thesis, we address the problem of Fine-grained Incident Video Retrieval (FIVR) using video similarity learning methods. FIVR is a video retrieval task that aims to retrieve all videos that depict the same incident given a query video { related video retrieval tasks adopt either very narrow or very broad scopes, considering only nearduplicate or same event videos. To formulate the case of same incident videos, we de ne three video associations taking into account the spatio-temporal spans captured by video pairs. To cover the benchmarking needs of FIVR, we construct a large-scale dataset, called FIVR-200K, consisting of 225,960 YouTube videos from major news events crawled from Wikipedia. The dataset contains four annotation labels according to FIVR de nitions; hence, it can simulate several retrieval scenarios with the same video corpus. To address FIVR, we propose two video-level approaches leveraging features extracted from intermediate layers of Convolutional Neural Networks (CNN). The rst is an unsupervised method that relies on a modi ed Bag-of-Word scheme, which generates video representations from the aggregation of the frame descriptors based on learned visual codebooks. The second is a supervised method based on Deep Metric Learning, which learns an embedding function that maps videos in a feature space where relevant video pairs are closer than the irrelevant ones. However, videolevel approaches generate global video representations, losing all spatial and temporal relations between compared videos. Therefore, we propose a video similarity learning approach that captures ne-grained relations between videos for accurate similarity calculation. We train a CNN architecture to compute video-to-video similarity from re ned frame-to-frame similarity matrices derived from a pairwise region-level similarity function. The proposed approaches have been extensively evaluated on FIVR- 200K and other large-scale datasets, demonstrating their superiority over other video retrieval methods and highlighting the challenging aspect of the FIVR problem.

To accommodate the growing demand for video streaming, the concept of Scalable Video Coding (SVC), also referred to as layered video coding, became the center of research efforts, which generates multiple inter-dependent video layers, including a Based Layer (BL) and multiple Enhancement Layers (ELs). Due to the fact that the ELs are encoded and decoded with reference to the BL, the decodability of the ELs is directly conditioned on that of the BL. This requirement benefits from Unequal Error Protection (UEP) techniques in terms of improving the attainable video quality by offering the strongest protection for the most important video bits. In this treatise, we mainly focus our attention on the physical layer based UEP techniques, including physical layer Forward Error Correction (FEC) and modulation techniques. Specifically, we consider a pair of FEC techniques, namely Inter-Layer FEC (IL-FEC) and variable-rate FEC schemes. Additionally, physical layer based UEP can also be achieved with the aid of modulation techniques. For example, the Hierarchical Quadrature Amplitude Modulation (HQAM) scheme is capable of providing simple UEP by mapping the more important bits to the specific constellation points having a larger Euclidean distance. Similarly, typical UEP relying on Multiple-Input Multiple-Output (MIMO) schemes feeds the more important bits to the specific MIMO subchannel experiencing better channel quality. Against the aforementioned background, in this thesis, we design physical layer based adaptive UEP for layered video streaming. Firstly, we exploit modulation based UEP by invoking optimal power-sharing assisted Superposition Coding (SC) schemes for video broadcasting services, where the bits of the various video layers are conveyed by different modulation modes relying on optimized power in order to maximize the overall video performance. Additionally, the UEP potential of the Index Modulation (IM) is investigated relying on a sophisticated Multi-Set Space-Time Shift Keying (MS-STSK), which provides a bit mapping assisted UEP scheme by feeding the source bits into the MS-STSK subchannels according to their importance. Moreover, the family of channel FEC based UEP techniques is also discussed. Specifically, we modify the conventional IL-FEC technique to provide UEP for the ELs, where an adaptive IL-FEC scheme is proposed for providing a gracefully decaying video performance erosion in the face of channel quality degradation. Finally, a variable FEC code-rate scheme is conceived for scalable panoramic video streaming.

In recent years, topic group mining and massive crowd opinion analysis from on-line social network platforms have become some of the most important tasks not only in research area but also in industry. Systems of this sort can identify similar topics from a very large dataset, group them together based on the topic, and analyse the inclination of the content's owner. To solve this problem, which involves research from a number of different areas, an integrated platform needs to be proposed. Most community mining techniques treat the network as a graph where nodes represent users and edges reflect user relationship between two users. One obvious drawback of these approaches is that it can only utilise the explicit user relationships provided by on-line social network platforms. All other possible relationships will be ignored. Some on-line social network platforms restrict the length of content a user can publish. This causes traditional document clustering methods to perform poorly. Meanwhile, the restriction of content length also affects opinion mining performance since most content lacks contextual features. Hence, other context features that are not immediately or obviously related need to be investigated to improve performance in user inclination classification. This research proposes a novel three layered platform. Two core technologies of the platform are topic group mining and user inclination analysis. The integrated approach was evaluated by a series of experiments to examine each core technology. The results indicate that the proposed integrated platform is able to produce the following results. 1) Scores up to 0.82 by V-measure evaluation function in topic group mining. 2) High accuracy rate in inclination mining. 3) A flexible and adaptable platform design which can accommodate different on-line social networks easily.

The focus of this thesis is twofold: on one hand we study the query complexity of equilibrium computation in games, and on the other hand, we use equilibrium concepts from game theory as a tool to understand miner incentives in Bitcoin. In terms of query complexity, we mostly focus on algorithms that have access to utility queries in large games and best response queries in bimatrix games. For the former, we demonstrate query-efficient completely uncoupled dynamics that achieve non-trivial approximate equilibria. For the latter, we reduce the problem of query-efficient approximate equilibrium computation to a natural geometric learning problem: approximately learning partitions of an n-dimensional simplex into disjoint convex polytopes via membership queries. Given this reduction we show query-efficient algorithms for the geometric problem, and ultimately provide an algorithm for computing ε-well-supported Nash equilibria in mxn bimatrix games with a query cost that is polynomial in log(1/ε) and max (m,n) provided that min (m,n) is constant. This leads to a polynomial query complexity algorithm for 2-player games,provided that one of the players has a constant number of strategies. As for incentives in Bitcoin, we shed some light into how robust honest mining protocols are to the presence of strategic agents. Our focus is on the strategic aspects of both solo mining and pool mining in Bitcoin. For the former, we take a multiplayer approach and exhibit specific strategy profiles of multiple strategic miners that outperform honest mining, even if said miners would not be incentivised to be dishonest individually. This effectively renders the Bitcoin protocol less secure than previously thought. As for the latter, we propose a new mining pool protocol that is a randomised variant of the already-ubiquitous pay-per-last-N-shares (PPLNS) mining pool scheme in Bitcoin. Our pool protocol, randomised pay-per-last-N-shares (RPPLNS),enjoys the same desirable properties of PPLNS, but with the added benefit of an exponentially reduced state space required to maintain the protocol. More importantly, this reduced state space also allows us to prove robust guarantees against a richer class of strategic pool mining than before.

Multimedia services consumption has increased tremendously since the deployment of 4G/LTE networks. Mobile video services (e.g., YouTube and Mobile TV) on smart devices are expected to continue to grow with the emergence and evolution of future networks such as 5G. The end user’s demand for services with better quality from service providers has triggered a trend towards Quality of Experience (QoE) - centric network management through efficient utilization of network resources. However, existing network technologies are either unable to adapt to diverse changing network conditions or limited in available resources. This has posed challenges to service providers for provisioning of QoE-centric multimedia services. New networking solutions such as Software Defined Networking (SDN) and Network Function Virtualization (NFV) can provide better solutions in terms of QoE control and management of multimedia services in emerging and future networks. The features of SDN, such as adaptability, programmability and cost-effectiveness make it suitable for bandwidth-intensive multimedia applications such as live video streaming, 3D/HD video and video gaming. However, the delivery of multimedia services over SDN/NFV networks to achieve optimized QoE, and the overall QoE-centric network resource management remain an open question especially in the advent development of future softwarized networks. The work in this thesis intends to investigate, design and develop novel approaches for QoE-centric control and management of multimedia services (with a focus on video streaming services) over software defined and virtualized networks. First, a video quality management scheme based on the traffic intensity under Dynamic Adaptive Video Streaming over HTTP (DASH) using SDN is developed. The proposed scheme can mitigate virtual port queue congestion which may cause buffering or stalling events during video streaming, thus, reducing the video quality. A QoE-driven resource allocation mechanism is designed and developed for improving the end user’s QoE for video streaming services. The aim of this approach is to find the best combination of network node functions that can provide an optimized QoE level to end-users through network node cooperation. Furthermore, a novel QoE-centric management scheme is proposed and developed, which utilizes Multipath TCP (MPTCP) and Segment Routing (SR) to enhance QoE for video streaming services over SDN/NFV-based networks. The goal of this strategy is to enable service providers to route network traffic through multiple disjointed bandwidth-satisfying paths and meet specific service QoE guarantees to the end-users. Extensive experiments demonstrated that the proposed schemes in this work improve the video quality significantly compared with the state-of-the- art approaches. The thesis further proposes the path protections and link failure-free MPTCP/SR-based architecture that increases survivability, resilience, availability and robustness of future networks. The proposed path protection and dynamic link recovery scheme achieves a minimum time to recover from a failed link and avoids link congestion in softwarized networks.

Detecting and segmenting Spatio-temporal foreground objects from videos are significant to motion pattern modelling and video content analysis. Extensive efforts have been made in the past decades. Nevertheless, video-based saliency detection and foreground segmentation remained challenging. On the one hand, the performances of image-based saliency detection algorithms are limited in complex contents, while the temporal connectivity between frames are not well-resolved. On the other hand, compared with the prosperous image-based datasets, the datasets in video-level saliency detection and segmentation usually have smaller scale and less diversity of contents. Towards a better understanding of video-level semantics, this thesis investigates the foreground estimation and segmentation in both image-level and video-level. This thesis firstly demonstrates the effectiveness of traditional features in video foreground estimation and segmentation. Motion patterns obtained by optical flow are utilised to draw coarse estimations about the foreground objects. The coarse estimations are refined by aligning motion boundaries with actual contours of the foreground objects with the participation of HOG descriptor. And a precise segmentation of the foreground is computed based on the refined foreground estimations and video-level colour distribution. Second, a deep convolutional neural network (CNN) for image saliency detection is proposed, which is named HReSNet. To improve the accuracy of saliency prediction, an independent feature refining network is implemented. A Euclidean distance loss is integrated into loss computation to enhance the saliency predictions near the contours of objects. The experimental results demonstrate that our network obtains competitive results compared with the state-of-art algorithms. Third, a large-scale dataset for video saliency detection and foreground segmentation is built to enrich the diversity of current video-based foreground segmentation datasets. A supervised framework is also proposed as the baseline, which integrates our HReSNet, Long-Short Term Memory (LSTM) networks and a hierarchical segmentation network. Forth, in the practice of change detection, there requires distinguishing the expected changes with semantics from the unexpected changes. Therefore, a new CNN design is proposed to detect changes in multi-temporal high-resolution urban images. Experimental results showed our change detection network outperformed the competing algorithms with significant advantages.

In this thesis, we present a qualitative exploration of how video bloggers (vloggers) achieve audience engagement in their vlogging. Engagement is a crucial factor for developing relationships between content providers and consumers, and for the quality of consumers' experience with media products such as online videos. Previous research in marketing and media studies suggests that achieving engagement with their audience is significant for the success of vloggers' careers, such as obtaining large fan bases on YouTube and other social media sites. However, how vloggers achieve engagement is rarely explored, and their audience engagement has rarely been clearly defined. We view vlog audience engagement as having two facets: consumer engagement and media engagement. This thesis applies theories of both customer and media engagement to the exploration of vlog consumption. By investigating three factors of vlogs - content, context, and presence - we address the question of how vloggers achieve audience engagement. We argue these three elements to be key to constructing audience engagement. We employ a case study research with a qualitative content analysis method applied to video samples from four well-known YouTube vloggers. Our results suggest that vlog audience engagement can be addressed through both customer and media engagement theories, which can help to account for vlogger-audience relationships beyond video consumption and audience experience during video viewing. To obtain consumer engagement, vloggers can use interaction, self-presentation, information management, and rewards in their vlog content. Vloggers' execution of these engagement factors in their content can also be conditioned by the vlogging context, namely the context of vlogger, audience, physical environment, social environment, and medium. Furthermore, sensory and social elements may evoke aspects of presence to contribute to media engagement. Sensory elements are colour, camera usage, image quality, and audio quality; social elements are the vlogger's eye contact, facial expressions, conversations and gestures. By applying both customer and media engagement theories to explore vlog audience engagement, this thesis bridges gaps regarding the application of engagement theories explicitly to vlogging. It indicates that vlogs are a useful medium to encourage audience engagement both inside and beyond vlog consumption. Our results suggest that customer engagement factors can be added to vlog content to encourage vlogger-audience connections beyond video consumption. We also indicate that context can influence the production of related customer engagement factors in vlog content. Using presence determinants in video production may affect vlog viewers' direct consumption experience for media engagement. This thesis provides a way to substantially understand vlog audience engagement by considering both viewers' video consumption experiences and the building of vlogger-audience relationships. The thesis also offers new directions for researchers to explore audience experience with vloggers from an engagement perspective through vlog content, context, and presence. Furthermore, our results could also be developed into practical guides in the future for new or existing vloggers to build and maintain connections with their audience and improve the latter's video consumption experience.

Low-latency multimedia applications generate a growing and significant majority of all Internet traffic. These applications are characterised by tight bounds on end-to-end latency that typically range from tens to a few hundred milliseconds. Operating within these bounds is challenging, with the best-effort delivery service of the Internet giving rise to unreliable delivery with unpredictable latency. The way in which the upper layers of the protocol stack manage this unreliability and unpredictability can greatly impact the quality-of-experience that applications can provide. In this thesis, I focus on the services and abstractions that the transport layer provides to applications. The delivery model provided by the transport layer can have a significant impact on the quality-of-experience that can be provided by the application. Reliability and order, for example, introduce delay while packet loss is detected and the lost data retransmitted. This enforces a particular trade-off between latency, loss, and application quality-of-experience, with reliability taking priority. This trade-off is not suitable for low-latency multimedia applications, which prefer predictable and bounded latency to strict reliability and order. No widely-deployed transport protocol provides a delivery model that fully supports low-latency applications: UDP provides no reliability guarantees, while TCP enforces reliability. Implementing a protocol that does support these applications is difficult: ossification restricts protocols to appearing as UDP or TCP on-the-wire. To meet both challenges -- of better supporting low-latency multimedia applications, and of deploying a new protocol within an ossified transport layer -- I propose TCP Hollywood, a protocol that maintains wire compatibility with TCP, while exposing the trade-off between reliability and delay such that applications can improve their quality-of-experience. I show that TCP Hollywood is deployable on the public Internet, and that it achieves its goal of improving support for low-latency multimedia applications. I conclude by evaluating the API changes that are required to support TCP Hollywood, distilling the protocol into the set of transport services that it provides.

Recently, the use of Twitter data has become important for a wide range of real-time applications, including real-time event detection, topic detection or disaster and emergency management. These applications require to know the precise location of the tweets for their analysis. However, approximately 1% of the tweets are finely-grained geotagged, which remains insufficient for such applications. To overcome this limitation, predicting the location of non-geotagged tweets, while challenging, can increase the sample of geotagged data to support the applications mentioned above. Nevertheless, existing approaches on tweet geolocalisation are mostly focusing on the geolocation of tweets at a coarse-grained level of granularity (i.e., city or country level). Thus, geolocalising tweets at a fine-grained level (i.e., street or building level) has arisen as a newly open research problem. In this thesis, we investigate the problem of inferring the geolocation of non-geotagged tweets at a fine-grained level of granularity (i.e., at most 1 km error distance). In particular, we aim to predict the geolocation where a given tweet was generated using its text as a source of evidence. This thesis states that the geolocalisation of non-geotagged tweets at a fine-grained level can be achieved by exploiting the characteristics of the 1\% of already available individual finely-grained geotagged tweets provided by the Twitter stream. We evaluate the state-of-the-art, derive insights on their issues and propose an evolution of techniques to achieve the geolocalisation of tweets at a fine-grained level. First, we explore the existing approaches in the literature for tweet geolocalisation and derive insights on the problems they exhibit when adapted to work at a fine-grained level. To overcome these problems, we propose a new approach that ranks individual geotagged tweets based on their content similarity to a given non-geotagged. Our experimental results show significant improvements over previous approaches. Next, we explore the predictability of the location of a tweet at a fine-grained level in order to reduce the average error distance of the predictions. We postulate that to obtain a fine-grained prediction a correlation between similarity and geographical distance should exist, and define the boundaries were fine-grained predictions can be achieved. To do that, we incorporate a majority voting algorithm to the ranking approach that assesses if such correlation exists by exploiting the geographical evidence encoded within the Top-N most similar geotagged tweets in the ranking. We report experimental results and demonstrate that by considering this geographical evidence, we can reduce the average error distance, but with a cost in coverage (the number of tweets for which our approach can find a fine-grained geolocation). Furthermore, we investigate whether the quality of the ranking of the Top-N geotagged tweets affects the effectiveness of fine-grained geolocalisation, and propose a new approach to improve the ranking. To this end, we adopt a learning to rank approach that re-ranks geotagged tweets based on their geographical proximity to a given non-geotagged tweet. We test different learning to rank algorithms and propose multiple features to model fine-grained geolocalisation. Moreover, we investigate the best performing combination of features for fine-grained geolocalisation. This thesis also demonstrates the applicability and generalisation of our fine-grained geolocalisation approaches in a practical scenario related to a traffic incident detection task. We show the effectiveness of using new geolocalised incident-related tweets in detecting the geolocation of real incidents reports, and demonstrate that we can improve the overall performance of the traffic incident detection task by enhancing the already available geotagged tweets with new tweets that were geolocalised using our approach. The key contribution of this thesis is the development of effective approaches for geolocalising tweets at a fine-grained level. The thesis provides insights on the main challenges for achieving the fine-grained geolocalisation derived from exhaustive experiments over a ground truth of geotagged tweets gathered from two different cities. Additionally, we demonstrate its effectiveness in a traffic incident detection task by geolocalising new incident-related tweets using our fine-grained geolocalisation approaches.

Cloud-based software services have become more popular and dependable and are ideal for businesses with growing or changing workload demands. These services are increasing rapidly due to the reduced hosting costs and the increased availability and efficiency of computing resources. The delivery of cloud-based software services is based on the underlying cloud infrastructure supported by cloud providers, which delivers the potential for scalability that follows the pay-as-you-go model. Performance and scalability testing and measurements of those services are necessary for future optimisations and growth of cloud computing to support the Service Level Agreement (SLA) compliant quality of cloud services, especially in the context of rapidly expanding quantity of service delivery. This thesis addresses an important issue, understanding the scalability of cloud-based software services from a technical perspective, which is very important as more software solutions are migrated to the cloud. A novel testing and quantifying approach for the scalability performance of cloud-based software services is described. Two technical scalability metrics for software services that have been deployed and distributed in cloud environments, have been formulated: volume and quality scalability metrics based on the number of software instances and the average response time. The experimental analysis comprises three stages. The first stage involves demonstrating the approach and the metrics using real-world could-based software service running on Amazon EC2 cloud using three demand scenarios. The second stage aims to extend the practicality of the metrics with experiments on two public cloud environments (Amazon EC2 and Microsoft Azure) with two cloud-based software serices to demonstrate the use of these metrics. The experimental analysis considers three sets of comparisons to provide the platform to construct the metrics as a basis that can be used effectively to compare the scalability of software on cloud environments, consequently supporting deployment decisions with technical arguments. Moreover, the work integrates the technical scalability metrics with an earlier utility-oriented scalability metric. The third stage is a case study of application-level fault inection using real-world cloud-based software services running on Amazon EC2 cloud to demonstrate the effect of fault scenarios on the scalability behaviour. The results show that the technical metrics quantify explicitly the technical scalability performance of the cloud-based software services, and that they allow clear assessment of the impact of demand scenarios, cloud platform and fault injection on the software services' scalability behaviour. The studies undertaken in this thesis have provided a valuable insight into the scalability of cloud-based software services delivery.

Social media platforms have today become an important part of people’s lives. People share their daily activities and interests through these platforms and also read about others’ activities. The increasing use of social media has led to rapid growth in the amount of shared information, and consequently has caused the challenge of information overload. This causes difficulties for users in filtering this huge amount of shared data in order to find and extract the information they need. This consumes the user’s time, increases information clutter and reduces their satisfaction. To manage this challenge, there is a need for personalisation services such as recommender systems. One popular social networking site is Twitter; it provides a rich environment for shared information that can help with recommender system research. However, short-text-based recommender systems based on Twitter activity suffer from a lack of reliable data. One solution to this problem is to build more powerful user profiles that reflect the recent interests of the user and consequently provide more accurate recommendations. A prompt way of building such a profile is to gather the user’s recent activities from their Twitter timeline. However, many users do not provide enough up to date data to build such a profile. Several researchers have tried to overcome this problem by enriching the profile via external textual sources such as Wikipedia. However, external sources may not be able to provide valid data that reflect the user’s interests and consequently the recommendation quality can be affected. This research studies short-text social media user modelling by utilising explicit and implicit relationships. This approach aims to build personal profiles through a different method, using tweets from the user’s Twitter network to provide more accurate recommendations. Our method exploits a Twitter user’s explicit and implicit relationships in order to extract other users’ important tweets that will help in building the personal profile. These relationships were identified by our proposed influence algorithm, which is a new way of measuring the influence rule from the user’s perspective rather than the influencers’ perspective. The usefulness of this proposed method is validated by implementing a tweet recommendation service based on the content-based recommender system mechanism, and by performing offline evaluation on a real dataset of 40 users collected from Twitter. The proposed user profiles are compared against other profiles, used as a baseline, based on similarity and distance metrics to check the effectiveness of the proposed method. Our proposed method shows increased performance gains in reflecting the user’s interests by recommending more accurate items. The main contribution of this work is an influence algorithm and a framework that is able to identify influential friends of a user (people that the user follows) more accurately than other influence and similarity algorithms, and then use them as sources in enriching the user’s profile via their tweets. It also helped us to enrich the profile with tweets that were collected from other sources, which were found via explicit and implicit relationships. As a result, more accurate recommendations can be delivered.

In a linear hypermedia system, all users are offered a standard series of hyperlinks. Adaptive Hypermedia (AH) tailors what the user sees to the user's goals, abilities, interests, knowledge and preferences. Adaptive Hypermedia is said to be the answer to the 'lost in hyperspace' phenomenon, where the user has too many hyperlinks to choose from, and has little knowledge to select the most appropriate hyperlink. AH offers a selection of links and content that is most appropriate to the current user. In an Adaptive Educational Hypermedia (AEH) course, a student's learning experiences can be personalised using a User Model (UM), which could include information such as the student's knowledge level, preferences and culture. Beside these basic components, a Goal Model (GM) can represent the goal the users should meet and a Domain Model (DM) would represent the knowledge domain. Adaptive strategies are sets of adaptive rules that can be applied to these models, to allow the personalisation of the course for students, according to their needs. From the many interacting elements, it is clear that the authoring process is a bottleneck in the adaptive course creation, which needs to be improved in terms of interoperability, usability and reuse of the adaptive behaviour (strategies). Authoring of Adaptive Hypermedia is considered to be difficult and time consuming. There is great scope for improving authoring tools in Adaptive Educational Hypermedia system, to aid already burdened authors to create adaptive courses easily. Adaptation specifications are very useful in creating adaptive behaviours, to support the needs of a group of learners. Authors often lack the time or the skills needed to create new adaptation specifications from scratch. Creating an adaptation specification requires the author to know and remember the programming language syntax, which places a knowledge barrier for the author. LAG is a complete and useful programming language, which, however, is considered too complex for authors to deal with directly. This thesis thus proposes a visual framework (LAGBlocks) for the LAG adaptation language and an authoring tool (VASE) to utilise the proposed visual framework, to create adaptive specifications, by manipulating visual elements. It is shown that the VASE authoring tool along with the visual framework enables authors to create adaptive specifications with ease and assist authors in creating adaptive specifications which promote the "separation of concern". The VASE authoring tool offers code completeness, correctness at design time, and also allows for adaptive strategies to be used within other tools for adaptive hypermedia. The goal is thus to make adaptive specifications easier, to create and to share for authors with little or no programming knowledge and experience. This thesis looks at three aspects of authoring in adaptive educational hypermedia systems. The first aspect of the thesis is concerned with problems faced by the author of an adaptive hypermedia system; the second aspect is concerned with describing the findings gathered from investigating the previously developed authoring tools; and the final aspect of the thesis is concerned with the proposal, the implementation and the evaluation of a new authoring tool that improves the authoring process for authors with different knowledge, background and experience. The purpose of the new tool, VASE, is to enable authors to create adaptive strategies in a puzzle-building manner; moreover, the created adaptive strategies could be used within (are compatible with) other systems in adaptive hypermedia, which use the LAG programming language.

This thesis investigates the practices and labours of beauty vloggers (video bloggers) and the inequality sustained by these labours and practices. Beauty vloggers regularly produce videos on themes including hair, beauty, fashion and relationships for their own stable, selfcontained, branded YouTube channels. The literature on beauty vlogging has examined the presentations of individual vloggers. In this thesis, I problematise the conception of beauty vlogging as a solo endeavour, situating beauty vloggers in a wider vlogging industry, in the specific geographic context of the UK. Through the lens of feminist political economy, I ask how the organisational (macro) structures, in addition to (micro) frictional interactions between industry stakeholders co-produce beauty vloggers' symbolic content. Analysis is informed by a wider three-year ethnographic study of British 'A List' vloggers on YouTube, conducted between 2015-2018, drawing from the "messy web" of research sites (Postill & Pink, 2012: 125). Ethnography thus encompasses digital, and offline elements, that make up the complexity and embodied nature of long term analysis of spaces contingent to platforms. I also conducted semi-structured interviews with beauty vloggers and stakeholders. Through 'zooming out' from analysis of the independent beauty vlogger this thesis considers how YouTube's multisided markets, algorithms and their interpretations, the subjective decisions of talent management organisations, relationships with advertisers, authenticity discourses and alignment with existing creative industries and intermediaries all shaped the content that becomes visible on YouTube. iii Very few women are able to create a sustainable career through YouTube production, although YouTube is increasingly promoted as a pathway to creative employment. In this thesis I counter YouTube's self-definition as an 'open platform'. I argue that organisational structures and interactions between stakeholders assign visibility and reify existing lines of societal inequalities, in addition to rewarding the production of commercial and feminised content in the vlogging industry.

Today, distributed applications run at data centre and Internet scales, from intensive data analysis, such as MapReduce; to the dynamic demands of a worldwide audience, such as YouTube. The network is essential to these applications at both scales. To provide adequate support, we must understand the full requirements of the applications, which are revealed by the workloads. In this thesis, we study distributed system applications at different scales to enrich this understanding. Large-scale Internet applications have been studied for years, such as social networking service (SNS), video on demand (VoD), and content delivery networks (CDN). An emerging type of video broadcasting on the Internet featuring crowdsourced live video streaming has garnered attention allowing platforms such as Twitch to attract over 1 million concurrent users globally. To better understand Twitch, we collected real-time popularity data combined with metadata about the contents and found the broadcasters rather than the content drives its popularity. Unlike YouTube and Netflix where content can be cached, video streaming on Twitch is generated instantly and needs to be delivered to users immediately to enable real-time interaction. Thus, we performed a large-scale measurement of Twitchs content location revealing the global footprint of its infrastructure as well as discovering the dynamic stream hosting and client redirection strategies that helped Twitch serve millions of users at scale. We next consider applications that run inside the data centre. Distributed computing applications heavily rely on the network due to data transmission needs and the scheduling of resources and tasks. One successful application, called Hadoop, has been widely deployed for Big Data processing. However, little work has been devoted to understanding its network. We found the Hadoop behaviour is limited by hardware resources and processing jobs presented. Thus, after characterising the Hadoop traffic on our testbed with a set of benchmark jobs, we built a simulator to reproduce Hadoops job traffic With the simulator, users can investigate the connections between Hadoop traffic and network performance without additional hardware cost. Different network components can be added to investigate the performance, such as network topologies, queue policies, and transport layer protocols. In this thesis, we extended the knowledge of networking by investigated two widelyused applications in the data centre and at Internet scale. We (i) studied the most popular live video streaming platform Twitch as a new type of Internet-scale distributed application revealing that broadcaster factors drive the popularity of such platform, and we (ii) discovered the footprint of Twitch streaming infrastructure and the dynamic stream hosting and client redirection strategies to provide an in-depth example of video streaming delivery occurring at the Internet scale, also we (iii) investigated the traffic generated by a distributed application by characterising the traffic of Hadoop under various parameters, (iv) with such knowledge, we built a simulation tool so users can efficiently investigate the performance of different network components under distributed application.

The methodology of intermediary XML schemas is introduced and its application to complex metadata environments is explored. Intermediary schemas are designed to mediate to other ‘referent’ schemas: instances conforming to these are not generally intended for dissemination but must usually be realized by XSLT transformations for delivery. In some cases, these schemas may also generate instances conforming to themselves. Three subsidiary methods of this methodology are introduced. The first is application-specific schemas that act as intermediaries to established schemas which are problematic by virtue of their over-complexity or flexibility. The second employs the METS packaging standard as a template for navigating instances of a complex schema by defining an abstract map of its instances. The third employs the METS structural map to define templates or conceptual models from which instances of metadata for complex applications may be realized by XSLT transformations. The first method is placed in the context of earlier approaches to semantic interoperability such as crosswalks, switching across, derivation and application profiles. The second is discussed in the context of such methods for mapping complex objects as OAI-ORE and the Fedora Content Model Architecture. The third is examined in relation to earlier approaches to templating within XML architectures. The relevance of these methods to contemporary research is discussed in three areas: digital ecosystems, archival description and Linked Open Data in digital asset management and preservation. Their relevance to future research is discussed in the form of suggested enhancements to each, a possible synthesis of the second and third to overcome possible problems of interoperability presented by the first, and their potential role in future developments in digital preservation. This methodology offers an original approach to resolving issues of interoperability and the management of complex metadata environments; it significantly extends earlier techniques and does so entirely within XML architectures.

Massive global video surveillance worldwide captures data but lacks detailed activity information to flag events of interest, while the human burden of monitoring video footage is untenable. Artificial intelligence (AI) can be applied to raw video footage to identify and extract required information and summarize it in linguistic formats. Video summarization automation usually involves text-based data such as subtitles, segmenting text and semantics, with little attention to video summarization in the processing of video footage only. Classification problems in recorded videos are often very complex and uncertain due to the dynamic nature of the video sequence and light conditions, background, camera angle, occlusions, indistinguishable scene features, etc. Video scene classification forms the basis of linguistic video summarization, an open research problem with major commercial importance. Soccer video scenes present added challenges due to specific objects and events with similar features (e.g. "people" include audiences, coaches, and players), as well as being constituted from a series of quickly changing and dynamic frames with small inter-frame variations. There is an added difficulty associated with the need to have light weight video classification systems working in real time with massive data sizes. In this thesis, we introduce a novel system based on Interval Type-2 Fuzzy Logic Classification Systems (IT2FLCS) whose parameters are optimized by the Big Bang-Big Crunch (BB-BC) algorithm, which allows for the automatic scenes classification using optimized rules in broadcasted soccer matches video. The type-2 fuzzy logic systems would be unequivocal to present a highly interpretable and transparent model which is very suitable for the handling the encountered uncertainties in video footages and converting the accumulated data to linguistic formats which can be easily stored and analysed. Meanwhile the traditional black box techniques, such as support vector machines (SVMs) and neural networks, do not provide models which could be easily analysed and understood by human users. The BB-BC optimization is a heuristic, population-based evolutionary approach which is characterized by the ease of implementation, fast convergence and low computational cost. We employed the BB-BC to optimize our system parameters of fuzzy logic membership functions and fuzzy rules. Using the BB-BC we are able to balance the system transparency (through generating a small rule set) together with increasing the accuracy of scene classification. Thus, the proposed fuzzy-based system allows achieving relatively high classification accuracy with a small number of rules thus increasing the system interpretability and allowing its real-time processing. The type-2 Fuzzy Logic Classification System (T2FLCS) obtained 87.57% prediction accuracy in the scene classification of our testing group data which is better than the type-1 fuzzy classification system and neural networks counterparts. The BB-BC optimization algorithms decrease the size of rule bases both in T1FLCS and T2FLCS; the T2FLCS finally got 85.716% with reduce rules, outperforming the T1FLCS and neural network counterparts, especially in the "out-of-range data" which validates the T2FLCSs capability to handle the high level of faced uncertainties. We also presented a novel approach based on the scenes classification system combined with the dynamic time warping algorithm to implement the video events detection for real world processing. The proposed system could run on recorded or live video clips and output a label to describe the event in order to provide the high level summarization of the videos to the user.

Research has identified that websites can gain an added advantage by getting listed in Search Engine results Pages (SERPs) during search sessions by searchers as SERPS refer targeted traffic to the websites. Search Engine Optimization (SEO) enables websites to become visible in search engines during search sessions for featured products or services. SEO is a complex technique which is directly affected by the ranking algorithms of search engines such as Google. Bearing in mind that Google employs in excess of 200 dynamic ranking factors in its algorithm it can be seen that optimization is not straightforward. Given this complex environment, websites find it difficult to initiate and implement SEO. SEO knowledge and experience gained from optimizing websites in the past is highly valuable and applicable to optimize websites both now and in the future. Therefore the main aim of the research in this thesis is to investigate the problem of optimization of websites using the prior experience gained through the optimization of several case study websites. To facilitate this, novel process models have been designed in order to capture the experience of implementing essential techniques of SEO and to explain the procedure of implementation of fundamental on-page SEO techniques that have been shown previously to yield results (i.e. increases in ranking) for past case study websites. Quantitative experiments and qualitative evaluation were undertaken to verify the efficacy of the novel process models through their application to case study websites. Mixed methods were used in order to answer the research questions, inductive experimental methods to produce, finesse and test the process models and qualitative enquiry through means of a focus group to gather peer review from professionals within the field who had previously been trained and conducted a trial using the process models. Implementation procedures of acknowledged essential on-page SEO techniques were identified from past case study websites, which have been represented in the novel process models designed in the current research and empirically investigated by applying them in the experimental case study websites. These models were applied through quantitative experiments that identified essential on-page SEO techniques which were then implemented in two experimental case study websites as per the procedures represented in the process models. These experiments have yielded positive results, resulting in establishing and/or enhancing the visibility of case study websites in SERPs. Further the implementation procedures of essential on-page SEO techniques were represented in the designed process models and stored in an SEO experience base on the principle of INRECA-II methodology. Results of the focus group suggest that the process models do achieve credible results (i.e. establishing and/or enhancing visibility of websites in SERPs) through their application and are suitable for use by both novices and professionals alike. Overall the results achieved from both the quantitative experiments and qualitative evaluation provide promising support to validate the created knowledge.

Service-oriented computing is a popular technology that facilitates the development of large-scale distributed systems. However, the modular composition and flexible coordination of such applications still remains challenging for the following reasons: 1) the services are provided as loosely coupled black boxes that only expose their interfaces to the environment; 2) interacting services are not usually known in advance: web services are dynamically chosen to fulfil certain roles and are often replaced by services with a similar functionality; 3) the nature of the service-based application is decentralised. Loose coupling of web services is often lost when it comes to the construction of an application from services. The reason is that the object-oriented paradigm, which is widely used in the implementation of web services, does not provide a mechanism for service interface self-tuning. As a result, it negatively impacts upon the interoperability of web services. In this dissertation we present a formal method for automatic service configuration in the presence of subtyping, polymorphism, and flow inheritance. This is a challenging problem. On the one hand, the interface description language must be flexible enough to maintain service compatibility in various contexts without any modification to the service itself. On the other hand, the composition of interfaces in a distributed environment must be provably consistent. Our method is based on constraint satisfaction and Boolean satisfiability. First, we define a language for specifying service interfaces in a generic form, which is compatible with a variety of contexts. The language provides support for parametric polymorphism, Boolean variables, which are used to control dependencies between any elements of interface collections, and flow inheritance using extensible records and variants. We implemented the method as a constraint satisfaction solver. In addition to this, we present a protocol for interface configuration. It specifies a sequence of steps that leads to the generation of context-specific service libraries from generic services. Furthermore, we developed a toolchain that performs a complete interface configuration for services written in C++. We integrated support for flexible interface objects (i.e. objects that can be transferred in the application along with their structural description). Although the protocol relies solely on interfaces and does not take behaviour concerns into account, it is capable of finding discrepancies between input and output interfaces for simple stateful services, which only perform message synchronisation. Two running examples (a three buyers use-case and an image processing application) are used along the way to illustrate our approach. Our results seem to be useful for service providers that run their services in the cloud. The reason is twofold. Firstly, interfaces and the code behind them can be generic as long as they are sufficiently configurable. No communication between service designers is necessary in order to ensure consistency in the design. Instead, the interface correspondence in the application is ensured by the constraint satisfaction algorithm, which we have already designed. Secondly, the configuration and compilation of every service are separated from the rest of the application. This prevents source code leaks in proprietary software which is running in the cloud.

In our digital era, social networking sites are becoming increasingly relevant to multiple aspects of individuals’ lives, including personal and professional. A group of professionals who heavily rely on these platforms for commercial purposes is book authors, as they use them to connect with readers and promote their books. In their attempts to establish their professional identities in these platforms, they face multiple challenges that public self- presentation carries. Three studies were conducted for this dissertation giving both qualitative and quantitative insights to the ways authors use social networking sites. The findings resulted in the identification of design opportunities that could facilitate online identity management for commercial purposes. The first study surveyed 103 authors about their attitudes and behaviours regarding information sharing via their professional social networking. It found that authors engage in rich self-presentation concerning multiple life aspects but avoid sharing information of a personal nature. This finding led to the development of Layered Identity Theory that represents all identity aspects, both intimate and non-intimate, as layered with inner layers conveying more personal information. Another practice of authors is the management of multiple accounts in a single platform to support different life aspects. There are also authors who use pen names to avoid links between their author selves and their personally identifiable life aspects when publishing their books and managing their social networking. A novel hybrid self-presentation was identified, as there were authors who reported sharing real information about themselves but alter it to prevent their mass audiences from knowing their real name identities. Multiple challenges could rise for them, such as accidentally sharing content in wrong accounts or revealing personally identifiable information. The second study, an online investigation that was based on observing authors’ behaviours in social networking sites, provided insights about authors’ activities across platforms and the content they generate in the most popular ones. The study verified that authors cover a variety of topics about multiple life aspects, including family and relationship; however their sharing cannot be considered personal. This finding established the relevance of Layered Identity Theory to online information sharing. The study also found that authors form online networks and employ their relationships with other users to further establish their professional identities. At the same time, they use social networking sites for creative purposes, such as feedback collection and direct promotion. Indications that authors’ online identity management conveys underlying tensions were identified and discussed. These tensions involve identity incoherence across platforms, difficulties in managing audiences, blurred profiles, and the possibility of negative involvement of others in their self-presentation. The final study, which comprised of two co-design workshops, explored how authors relate to tensions that were previously identified in the studies conducted. In addition, the workshops uncovered authors’ experiences with other users of social networking sites who were negatively involved with their identity management. Managing multiple accounts in a single platform was confirmed by authors as a challenging activity, whereas sharing multiple life aspects was described as a positive common practice among authors when there is an overall balance in the types of information shared. Interestingly, authors’ motivation for sharing multiple life aspects with their online audiences is their desire to attract large audiences and be discovered by publishing companies. Authors acknowledge the underlying tensions that sharing multiple life aspects could generate, such as attracting audiences not interested in writing. When it comes to others’ involvement in their self- presentation, it was found that despite authors’ attempts to balance boundaries between their professional and other life aspects in social networking sites, the information fluidity that characterises the online world makes their different identities come together. Authors acknowledged not being fully aware of the different features that exist in social networking sites; however they had novel design ideas, such as prioritisation and categorisation of generated content, extra layers of security before content sharing, and greater control over others’ content about them. Overall, the findings of this dissertation reveal that even though social networking sites lack features to fully support authors’ self-presentation, the creativity that characterises this group of professionals allows them to use these platforms in ways that partially fulfil their needs; yet remain challenging and need to be facilitated. The studies conducted suggest that many opportunities rise for them online, such as the development of personal relationships with readers and other authors, despite the fact that their self- presentation is based on carefully planned activities. Moreover, the findings uncover that authors establish a new identity aspect online that is formed using a pen name and slightly altered information from their other identity aspects. By identifying Layered Identity Theory to precisely study information sharing online, hybrid self-presentation, as well as other commercial online practices, underlying tensions, and design recommendations, this dissertation raises considerations that could be used to further investigate the commercial use of social networking sites and design relevant supporting technology.

This thesis explores the challenges in abstract library specification, library refinement and reasoning about fine-grained concurrent programs. For abstract library specification, this thesis applies structural separation logic (SSL) to formally specify the behaviour of several libraries in an abstract, local and compositional manner. This thesis further generalises the theory of SSL to allow for library specifications that are language-independent. Most notably, we specify a fragment of the Document Object Model (DOM) library. This result is compelling as it significantly improves upon existing DOM formalisms in that the specifications produced are local, compositional and language-independent. Concerning library refinement, this thesis explores two existing approaches to library refinement for separation logic, identifying their advantages and limitations in different settings. This thesis then introduces a hybrid approach to refinement, combining the strengths of both techniques for simple scalable library refinement. These ideas are then adapted to refinement for SSL by presenting a JavaScript implementation of the DOM fragment studied and establishing its correctness with respect to its specification using the hybrid refinement approach. As to concurrent reasoning, this thesis introduces concurrent local subjective logic (CoLoSL) for compositional reasoning about fine-grained concurrent programs. CoLoSL introduces subjective views, where each thread is verified with respect to a customised local view of the state, as well as the general composition and framing of interference relations, allowing for better proof reuse.

Users of Location-Based Social Networks (LBSN) are giving away information about their whereabouts, and their interactions in the geographic space. In comparison to other types of personal data, location data are sensitive and can reveal user’s daily routines, activities, experiences and interests in the physical world. As a result, the user is facing an information overload that overburdens him to make a satisfied decision on where to go or what to do in a place. Thus, finding the matching places, users and content is one of the key challenges in LSBNs. This thesis investigates the different dimensions of data collected on LBSNs and proposes a user and place modelling framework. In particular, this thesis proposes a novel approach for the construction of different views of personal user profiles that reflect their interest in geographic places, and how they interact with geographic places. Three novel modelling frameworks are proposed, the static user model, the dynamic user model and the semantic place model. The static user model is a basic model that is used to represent the overall user interactions towards places. On the other hand, the dynamic user model captures the change of the user’s preferences over time. The semantic place model identifies user activities in places and models the relationships between places, users, implicit place types, and implicit activities. The proposed models demonstrate how geographic place characteristics as well as implicit user interactions in the physical space can further enrich the user profiles. The enrichment method proposed is a novel method that combines the semantic and the spatial influences into user profiles. Evaluation of the proposed methods is carried out using realistic data sets collected from the Foursquare LBSN. A new Location and content recommendation methods are designed and implemented to enhance existing location recommendation methods and results showed the usefulness of considering place semantics and the time dimension when the proposed user profiles in recommending locations and content. The thesis considers two further related problems; namely, the construction of dynamic place profiles and computing the similarity between users on LBSN. Dynamic place profiles are representations of geographic places through users’ interaction with the places. In comparison to static place models represented in gazetteers and map databases, these place profiles provide a dynamic view of how the places are used by actual people visiting and interacting with places on the LBSN. The different views of personal user profiles constructed within our framework are used for computing the similarity between users on the LBSN. Temporal user similarities on both the semantic and spatial levels are proposed and evaluated. Results of this work show the challenges and potential of the user data collected on LBSN.

Nowadays, a typical video content provider serves a variety of platforms e.g. smartphones, web browsers, and smart TVs. Each of these platforms has specific requirements with respect to transmission and video quality. Moreover, since these devices are increasingly being used on-the-go, the environment within which most of these video streaming clients operate is both unreliable and time-varying. To cater for these heterogeneous requirements, content providers are increasingly adopting adaptive streaming services. Through such services, the quality of the video content received by a user is adapted to fit its specific requirements and capabilities. To adapt the video quality, system capabilities such as network capacity and memory have to be continuously monitored and measured, chunk requests have to be scheduled, and then the optimal video rate has to be decided. Each of these tasks is usually managed by a sub-module of the adaptive bitrate selection function. However, these sub-components interact in a non-trivial manner. For example, while on-off chunk scheduling helps to prevent buffer overflow, it negatively affects the TCP throughput. Hence, these complex interactions between these different sub-components of the adaptive streaming algorithm result in unnecessary rebufferings, undesirable variability, and sub-optimal video quality. To help simplify these interactions, this thesis develops several frameworks and models that define the relationships between the various components of the adaptive bitrate selection system. This includes deriving the valid system state space, which defines the state that an algorithm can be in at any given time, determining the allowable interactions between the various components, and identifying the video quality evolution rules that optimise QoE. Using this information, some state-of-the-art algorithms are improved and novel ones developed to demonstrate the effectiveness of the proposed approach. The result of extensive evaluations conducted both within a real-world Internet environment and with network trace shows the proposed schemes help in reducing the convergence time, startup delay, and rebuffering events, while at the same time increasing both the average and the stability of the video quality. All this is obtained without any adverse impact on the fairness among the competing players.

This thesis proposes a reference model and its computational core to support the creation of software applications within educational environments, which address Differences In Learning (DiffInL) and are applicable to both learners and instructors. This work differs from others in that the strength of this model resides on the re-usable character of the reasoning mechanism enabled by the computational environment. The starting point is the definition of agreed learning goals that the learner needs to achieve. In turn, the reference model generates personalised, best-practice teaching and learning materials, suitable for achieving the individual’s learning goals. This reference model consists of MODEL and MANAGEMENT components. The MODEL components store the domain needed to create learners and instructional models, which are required for the creation of Learning Spaces (LeS). The MANAGEMENT compo- nent also manages the semantics stored in various model components in order to carry out the configuration of an LeS. The architecture of software applications generated from the reference model is illustrated and contains: Netbeans IDE 8.0.2, JavaServer Faces framework and OWL-API library. We tested this to generate teaching practices for Learning Difficulties (LDif) student. In order to prove the feasibility of creating a software application from the reference model, an example of a particular scenario in a specific educational setting for LDif Students has been shown. This proposed model has successfully proved its ability to address the needs of LDif Students through a corresponding novel and re-usable reasoning mechanism implemented in Web Ontology Language (OWL) and Semantic Web Rule Language (SWRL) computational environments. The reference model has shown its ability to integrate with different disciplines such as psychology, sociology and human-computer interactions. The main contribution to research is the creation of a novel reference computational model which addresses the needs of people with DiffInL. The strength of this model resides on the re-usable character of the reasoning mechanism enabled by the computational environment. The whole framework allows a unified implementation which takes into account classes, constraints, matching, and inference mechanisms for the complete configuration of an LeS. The suggested approach also differs from previous work in that it is personalised, and the applied reasoning rules are dynamic. Therefore this model can be constantly “tuned” according to the questions we may ask in such environments. Overall, the proposed reference model in this research offers a promising and feasible solution that can support current educational systems and benefit both learners and instructors. It also demonstrates the applicability of the latest technologies and would allow for future technologies to be incorporated, in order to enhance the model.

Culture affects the preferences of users and their expectations of the elements found in the user interfaces of programs. This study investigates the influence of culture, as well as the specific experience and expectations of Saudi Arabian users, with respect to the design of the user interface of the Arabic version of Facebook. As the Arabic version of Facebook has issues with the engagement of users, this research therefore sought to develop a deeper understanding of the causes of this and to propose viable solutions. This involved a comprehensive examination of the implications of translating user- interface elements from one language to another, looking in particular at the specific cultural needs of Saudi Arabian users, with the intention of improving users’ engagement. This study uses an interpretivist research based on the social construction that was utilised in data collection, consisting of interviews and two sequential questionnaires. The main issues identified in the design of the Facebook user interface were the icons and page layout. The current icon images were shown to be unfamiliar and difficult for Saudi users to understand. This was exacerbated by the labels used, which were also found to be too abstract to be meaningful. Regarding the page layout, the alignments were mixed and inconsistent, the positions of elements were horizontally displayed, the pages were too complicated for effective navigation, and the list of options was too long, adversely affecting comprehension. In addition, there were many language errors in the translated Arabic version. These issues led to confusion and overload among Saudi users, causing stress when navigating through the pages. These factors combined to contribute to a lack of engagement with the Arabic version among Saudi users. The consequences of this experience on users’ behaviour was also investigated. First, some users were found to confine their interaction with Facebook to the home page and to avoid the setting page. Many users expressed a preference for the English version of Facebook over the Arabic version, although some acclimated to the use of the Arabic version, while others migrated to other social media platforms. The findings show that application of the localisation theory demonstrates that user interfaces designed for one culture cannot always be easily modified to suit another culture without significant modification. Therefore, this research has contributed to the knowledge through the development of design guidelines to aid designers in the creation of a more suitable Arabic version of Facebook. Based on the above findings, this design guidelines proposes to guide user interface designers to incorporate elements that take into account cultural issues into the user interface.

With the emergence and growing popularity of online social networks, depth sensors (such as Kinect), smart phones /tablets, wearable devices, and augmented reality (such as Google Glass and Google Cardboard), the way in which people interact with digital media has been completely transformed. Globally, the apparel market is expected to grow at a compound annual growth rate of 5 between 2012 and 2025. Due to the huge impact for ecommerce applications, there is a growing interest in methods for clothing retrieval and outfit recommendation, especially efficient ones suitable for mobile apps. To this end, we propose a practical and efficient method for mobile visual clothing search and implement it as a smart phone app that enables the user to capture a photo of clothing of interest with their smart phone and retrieve similar clothing products that are available at nearby retailers. Furthermore, we propose an extended method where soft biometric clothing attributes are combined with anthropometrics computed from depth data for person identification and surveillance applications. This addresses the increased terrorist threat in recent years that has driven the need for non-intrusive person identification that can operate at a distance without a subject’s knowledge or collaboration. We implement the method in a wearable mobile augmented reality application based on a smart phone with Google Cardboard in order to demonstrate how a security guard could have their vision augmented to automatically identify a suspect in their field of vision. Lastly, we consider that a significant proportion of photos shared online and via apps are selfies and of dressed people in general. Hence, it is important both for consumers and for industry that systems are developed to understand the visual content in the vast datasets of networked content to aid management and perform smart analysis. To this end, this dissertation introduces an efficient technique to segment clothing in photos and recognize clothing attributes. We demonstrate with respect to the emerging augmented reality field by implementing an augmented reality mirror app for mobile tablet devices that can segment a user’s clothing in real-time and enable them to realistically see themselves in the mirror wearing variations of the clothing with different colours or graphics rendered. Empirical results show promising segmentation, recognition, and augmented reality performance.

The Web has provided a platform for people to collaborate by using collective intelligence. Messaging boards, Q&A forums are some examples where people broadcast their issues and other people provide solutions. Such communities are defined as a Purposive Social Network (PSN) in this thesis. PSN is a community where people with similar interest and varied expertise come together, use collective intelligence to solve common problems in the community and build tools for common purpose. Usually, Q&A forums are closed or semi-open. The data are controlled by the websites. Diculties in the search and discovery of information is an issue. People searching for answers or experts in a website can only see results from its own network, while losing a whole communityof experts in other websites. Another issue in Q&A forums is not getting any response from the community. There is a long tail of questions that get no answer. The thesis introduces the Suman system that utilises Semantic Web (SW) and Linked Data technologies to solve above challenges. SW technologies are used to structure the community data so it can be decentralized and used across platforms. Linked Data helps to find related information about linked resources. The Suman system uses available tools to solve name entity disambiguation problem and add semantics to the PSN data. It uses a novel combination of semantic keyword search with traditional text search techniquesto nd similar questions with answers for unanswered questions to expand thequery term with added semantics and uses corwdsourced data to rank the results. Furthermore, the Suman system also recommends experts who can answer those questions. This helps to narrow down the long tail of unanswered questions in such communities. The Suman system is designed using the Design Science methodology and evaluated by users in two experiments. The results were statistically analysed to show that the keywords generated by the Suman system were rated higher than the original keywords from the websites. It also showed that the participants agreed with the algorithm rating for answers provided by the Suman system. StackOverflow and Reddit are used as an example of PSN and to build an application as a proof of concept of the Suman system.

Due to their openness and low publishing barrier nature, User-Generated Content (UGC) platforms facilitate the creation of huge amounts of data, containing a substantial quantity of inaccurate content. The presence of misleading, questionable and inaccurate content may have detrimental effects on people's beliefs and decision-making and may create a public disturbance. Consequently, there is significant need to evaluate information coming from UGC platforms to differentiate credible information from misinformation and rumours. In this thesis, we present the need for research about online Arabic information credibility and argue that by extending the existing automated credibility assessment approaches to adding an extra step to evaluate labellers will lead to a more robust dataset for building the credibility classification model. This research focuses on modelling the credibility of Arabic information in the presence of disagreed judging credibility scores and ground truth of credibility information is not absolute. First, in order to achieve the stated goal, this study employs the idea of crowdsourcing whereby users can explicitly express their opinions about the credibility of a set of tweet messages. This information coupled with the data about tweets’ features enables us to identify messages’ prominent features with the highest usage in determining information credibility levels. Then experiments based on both statistical analysis using features’ distributions and machine learning methods are performed to predict and classify messages’ credibility levels. A novel credibility assessment model which integrates the labellers’ reliability weights is proposed when deriving the credibility labels for the messages in the training and testing dataset. This credibility model primarily uses similarity and accuracy rating measurements for evaluating the weighting of labellers. In order to evaluate proposed model, we compare the labelling obtained from the expert labellers with those from the weighted crowd labellers. Empirical evidence proposed that the credibility model is superior to the commonly used majority voting baseline compared to the experts’ rating evaluations. The observed experimental results exhibit a reduction of the effect of unreliable labellers’ credibility judgments and a moderate enhancement of the credibility classification results.

Machines that provide decision support have traditionally used either a representation of human expertise or used mathematical algorithms. Each approach has its own limitations. This study helps to combine both types of decision support system for a single system. However, the focus is on how the machines can formalise and manipulate the human representation of expertise rather than on data processing or machine learning algorithms. It will be based on a system that represents human expertise in a psychological format. The particular decision support system for testing the approach is based on a psychological model of classification that is called the Galatean model of classification. The simple classification problems only require one XML structure to represent each class and the objects to be assigned to it. However, when the classification system is implemented as a decision support system within more complex realworld domains, there may be many variations of the class specification for different types of object to be assigned to the class in different circumstances and by different types of user making the classification decision. All these XML structures will be related to each other in formal ways, based on the original class specification, but managing their relationships and evolution becomes very difficult when the specifications for the XML variants are text-based documents. For dealing with these complexities a knowledge representation needs to be in a format that can be easily understood by human users as well as supporting ongoing knowledge engineering, including evolution and consistency of knowledge. The aim is to explore how semantic web technologies can be employed to help the knowledge engineering process for decision support systems based on human expertise, but deployed in complex domains with variable circumstances. The research evaluated OWL as a suitable vehicle for representing psychological expertise. The task was to see how well it can provide a machine formalism for the knowledge without losing its psychological validity or transparency: that is, the ability of end users to understand the knowledge representation intuitively despite its OWL format. The OWL Galatea model is designed in this study to help in automatic knowledge maintenance, reducing the replication of knowledge with variant uncertainties and support in knowledge engineering processes. The OWL-based approaches used in this model also aid in the adaptive knowledge management. An adaptive assessment questionnaire is an example of it, which is dynamically derived using the users age as the seed for creating the alternative questionnaires. The credibility of the OWL Galatea model is tested by applying it on two extremely different assessment domains (i.e. GRiST and ADVANCE). The conclusions are that OWLbased specifications provide the complementary structures for managing complex knowledge based on human expertise without impeding the end users’ understanding of the knowledgebase. The generic classification model is applicable to many domains and the accompanying OWL specification facilitates its implementations.

Context/Background: Use of Service Level Agreements (SLAs) is crucial to provide the value added services to consumers to achieve their requirements successfully. SLAs also ensure the expected Quality of Service to consumers. Aim: This study investigates how efficient structural representation and management of SLAs can help to ensure the Quality of Service (QoS) in Web services during Web service composition. Method: Existing specifications and structures for SLAs for Web services do not fully formalize and provide support for different automatic and dynamic behavioral aspects needed for QoS calculation. This study addresses the issues on how to formalize and document the structures of SLAs for better service utilization and improved QoS results. The Service Oriented Architecture (SOA) is extended in this study with addition of an SLAAgent, which helps to automate the QoS calculation using Fuzzy Inference Systems, service discovery, service selection, SLA monitoring and management during service composition with the help of structured SLA documents. Results: The proposed framework improves the ways of how to structure, manage and monitor SLAs during Web service composition to achieve the better Quality of Service effectively and efficiently. Conclusions: To deal with different types of computational requirements the automation of SLAs is a challenge during Web service composition. This study shows the significance of the SLAs for better QoS during composition of services in SOA.

Context: Locating resources on the Web has become increasingly difficult for users and poses a number of issues. The sheer size of the Web means that despite what appears to be an increase in the amount of quality material available, the effort involved in locating that material is also increasing; in effect, the higher quality material is being diluted by the lesser quality. One such group affected by this problem is post-graduate students. Having only a finite amount of time to devote to research, this reduces their overall quality study time. Aim: This research investigates how post-graduate students use the Web as a learning resource and identifies a number of areas of concern with its use. It considers the potential for improvement in this matter by using a number of concepts such as: collaboration; peer reviewing and document classification and comparison techniques. This research also investigates whether by combining several of the identified technologies and concepts, student research on the Web can be improved. Method: Using some of the identified concepts as components, this research proposes a model to address the highlighted areas of concern. The proposed model, named the Durham Browsing Assistant (DurBA) is defined, and a number of key concepts which show potential within it are uncovered. One of the key concepts is chosen, that of document comparison. Given a source document, can a computer system reliably identify other documents which most closely match it from other on the Web? A software tool was created which allowed the testing of document comparison techniques, this was called the Durham Textual Comparison system (DurTeC) and it had two key concepts. The first was that it would allow various algorithms to be applied to the comparison process. The second concept was that it could simulate collaboration by allowing data to be altered, added and removed as if by multiple users. A set of experiments were created to test these algorithms and identify those which gave the best results. Results: The results from the experiments identified a number of the most promising relationships between comparison and collaboration processes. It also highlighted those which had a negative effect on the process, and those which produced variable results. Amongst the results, it was found that: 1. By providing DurTeC with additional source documents to the original, as if through a recommendation process, it was able to increase its accuracy substantially. 2. By allowing DurTeC to use synonym lists to expand its vocabulary, in many cases, it was found to have reduced its accuracy. 3. By restricting those words which DurTeC considered in its comparison process, based upon their value in the source document, accuracy could be increased. This could be considered as a form of collaborative keyword selection. Conclusion: This research shows that improvements can be made in the accuracy of identifying similar resources by using a combination of comparison and collaboration processes. The proposed model, DurBA would be an ideal host for such a system.

Continuous improvements to the state of the art have made it easier to create, send and receive vast quantities of video over the Internet. Catalysed by these developments, video is now the largest, and fastest growing type of traffic on modern IP networks. In 2015, video was responsible for 70% of all traffic on the Internet, with an compound annual growth rate of 27%. On the other hand, concerns about the growing energy consumption of ICT in general, continue to rise. It is not surprising that there is a significant energy cost associated with these extensive video usage patterns. In this thesis, I examine the energy consumption of typical video configurations during decoding (playback) and encoding through empirical measurements on an experimental test-bed. I then make extrapolations to a global scale to show the opportunity for significant energy savings, achievable by simple modifications to these video configurations. Based on insights gained from these measurements, I propose a novel, energy-aware Quality of Experience (QoE) metric for digital video - the Energy - Video Quality Index (EnVI). Then, I present and evaluate vEQ-benchmark, a benchmarking and measurement tool for the purpose of generating EnVI scores. The tool enables fine-grained resource-usage analyses on video playback systems, and facilitates the creation of statistical models of power usage for these systems. I propose GreenDASH, an energy-aware extension of the existing Dynamic Adaptive Streaming over HTTP standard (DASH). GreenDASH incorporates relevant energy-usage and video quality information into the existing standard. It could enable dynamic, energy-aware adaptation for video in response to energy-usage and user ‘green' preferences. I also evaluate the subjective perception of such energy-aware, adaptive video streaming by means of a user study featuring 36 participants. I examine how video may be adapted to save energy without a significant impact on the Quality of Experience of these users. In summary, this thesis highlights the significant opportunities for energy savings if Internet users gain an awareness about their energy usage, and presents a technical discussion how this can be achieved by straightforward extensions to the current state of the art.

The behavioural changes which result from expansion and popularity of online platforms can be observed in day to day activities such as buying tickets, ordering food, and communicating with friends and family. From a technical perspective these platforms utilise similar technologies and infrastructures (e.g. the Internet, servers, and databases), however there are numerous conceptual and functional differences which can be identified between them. This research is associated with specific types of online platforms which are commonly referred to as Web 2.0 platforms. Popular examples of such platforms include Wikipedia, YouTube, and Facebook. These platforms rely upon user generated content and benefit from users’ collaborative efforts. Users’ efforts can be considered as a resource, and a possible objective in associated projects could be to optimise the outcome of such efforts. In many cases, this can be accomplished by influencing users’ Online Behaviour (OB). In Web 2.0 platforms, OB can be influenced by various factors, and among them Sense of Belonging (SB) has been highlighted by many previous researchers as a significant factor. While the literature suggests a significant association between SB and OB in Web 2.0 platforms, this research identifies a gap in the existing literature in regards to SB due to its qualitative nature and the dependency of associated investigations on human interaction. In addition to the subjective nature of such investigations, other practical challenges can be associated with measuring SB in Web 2.0 projects. Firstly, considering the demographical, cultural and lingual diversity of users in many Web 2.0 platforms, such investigations can be complex and therefore expensive. Secondly, investigating SB cannot be performed in real time which can result in extended timelines for gathering data. Thirdly, in early stages of Web 2.0 projects in which the concept is being defined and the platform is being developed, the users might not yet exist, and hence investigating their SB and OB may not be an option. Finally, without an existing platform and community, there is nothing to belong to, and hence SB might not exist to measure or investigate. As an attempt to fill the identified gap and provide practical solutions for highlighted challenges in Web 2.0 projects, this research introduces a new factor, and investigates its association with OB. Virtual Belonging (VB) is proposed as a factor built upon suggested elements for SB, and based on a new perspective towards Web 2.0 features. Furthermore, in any Web 2.0 platform, VB can be quantitatively defined, compared, and analysed. This research suggests three dimensions for VB, and their associations with OB are hypothesised. To test the research hypotheses, a Web 2.0 platform was developed in association with a student poetry competition in the University of Manchester, and it was used as the vehicle for an experiment (available at epsPoetry.com). The platform includes a bespoke Tracking System that has been specifically designed and developed to assist with this investigation. Also, Ethical Approval was obtained from the University’s Ethics Committees for the use of the data collected by the Tracking System from the individuals who have agreed to take part in this research and used the platform between March 2013 and July 2014.VB’s dimensions are presented in this research as a hierarchy with three distinct levels, namely Inclusion, Involvement, and Influence. The results show a significant association between every dimension of users’ VB and their OB. On this basis, this research proposes VB as a predictive factor for OB in Web 2.0 projects. The findings can provide practical guidance and directions for Web 2.0 project managers, particularly in the early stages of such projects in which the existence of users and their interactivity with the platform can only be an assumption, and hence obtaining qualitative data (such as SB) may not be an option.

In an age when people are predisposed to report real-world events through their social media accounts, many researchers value the advantages of mining such unstructured and informal data from social media. Compared with the traditional news media, online social media services, such as Twitter, can provide more comprehensive and timely information about real-world events. Existing Twitter event monitoring systems analyse partial event data and are unable to report the underlying stories or sub-events in realtime. To ll this gap, this research focuses on the automatic identi cation of content for events and sub-events through the analysis of Twitter streams in real-time. To full the need of real-time content identification for events and sub-events, this research First proposes a novel adaptive crawling model that retrieves extra event content from the Twitter Streaming API. The proposed model analyses the characteristics of hashtags and tweets collected from live Twitter streams to automate the expansion of subsequent queries. By investigating the characteristics of Twitter hashtags, this research then proposes three Keyword Adaptation Algorithms (KwAAs) which are based on the term frequency (TF-KwAA), the tra c pattern (TP-KwAA), and the text content of associated tweets (CS-KwAA) of the emerging hashtags. Based on the comparison between traditional keyword crawling and adaptive crawling with di erent KwAAs, this thesis demonstrates that the KwAAs retrieve extra event content about sub-events in real-time for both planned and unplanned events. To examine the usefulness of extra event content for the event monitoring system, a Twitter event monitoring solution is proposed. This \Detection of Sub-events by Twit- ter Real-time Monitoring (DSTReaM)" framework concurrently runs multiple instances of a statistical-based event detection algorithm over different stream components. By evaluating the detection performance using detection accuracy and event entropy, this research demonstrates that better event detection can be achieved with a broader coverage of event content.

The rapid growth and influence of social network services has led many scholars to focus on privacy issues. However, the research described in this thesis was motivated by the small number of design studies that have focused on practical approaches to identifying tacit information from users’ instant non-verbal responses to privacy issues. The research therefore aimed to propose persona models as a design solution for software agent development based on the analysis of users’ emotional and behavioural responses, so as to address privacy issues in social network services. In the definition phase, 21 stakeholders belonging to three key stakeholder groups were recruited for unstandardised semistructured email interviews. Three main considerations for the design of software agents in social network services emerged from the interviews, which were classified into the following categories: comprehensive understanding of users’ perceptions of privacy; user type recognition algorithm for software agent development; and existing software agent enhancement. In the development phase, 50 participants were recruited for the Facebook case study, which included three research strategies: perceptions of privacy questionnaire for user typology; emotional response measurement using Geneva Emotion Wheel; and behavioural response observation using a contextual inquiry method. The participants were classified into four user types by means of cluster analysis: uninformed, trustful, suspicious and neglectful. Each user type’s key emotional responses were identified using Kruskal-Wallis test and Mann-Whitney U test, and key behavioural responses using affinity diagrams. This generated persona models of each user type that reflected the correlations between users’ perceptions of privacy, key emotional responses and key behavioural responses. Two fundamental features of the software agent were also proposed based on the persona models: confirmation and guidance. In the validation phase, software agent prototypes were created based on the proposed persona models. A total of 206 participants completed the online survey which included two sections: perceptions of privacy questionnaire for user typology replication, and key emotional responses measurement before and after the intervention of the software agent prototypes. Cluster analysis replication validated the proposed user typology, and Wilcoxon signed-rank test of key emotional responses validated the proposed persona models. By implementing the research outcomes, the software agent described in this thesis would be able to provide users with appropriate services based on their user types, to reduce the number of those who are still unaware of privacy practice and those who neglect their accounts, and to expand the size of a user group pursuing sound relationships.

The significant growth in multimedia services and traffic (e.g. VoIP, video streaming and video gaming) in current and emerging mobile networks including the latest 4G Long-Term Evolution (LTE) networks and the rising user expectation for high Quality of Experience (QoE) for these services have posed real challenges to network operators and service providers. One of the key challenges is how to bring multimedia services to the end-user over resource-constrained mobile networks with a satisfactory QoE. Cost-effective solutions are needed for network operators to improve the bandwidth usage of these mobile networks. Therefore, scheduling schemes are of extreme importance in LTE, where scheduling algorithms are responsible for the overall efficiency of resource allocation in an LTE system. The aim of the project is to develop novel QoE-driven scheduling algorithms for improving system capacity in delivering multimedia services over downlink 3GPP LTE. This is to move away from traditional QoS-driven scheduling schemes to a QoE-driven scheme which guarantee end-user satisfaction in resource allocation. The main contributions of the thesis are threefold: 1. Performance of several existing scheduling algorithms for VoIP applications was evaluated thoroughly in terms of QoE metric (i.e. MOS), instead of QoS metrics (e.g. packet loss and delay). Using QoE metrics instead of QoS ones will facilitate the development of QoE-driven scheduling schemes in order to achieve optimised end-user experiences or optimised mobile system capacity. 2. A novel QoE-driven LTE downlink scheduling scheme for VoIP application was developed to maximize the number of users per cell at an acceptable MOS score. The proposed scheme achieved significant improvement in cell capacity at an acceptable quality (75% compared to MLWDF, and 250% compared to PF and EXP-PF in all three lower speed scenarios considered). 3. A QoE-driven LTE downlink scheduling scheme for multiservice multimedia applications was developed to improve the cell capacity with satisfactory QoE for both VoIP and video streaming services. The proposed algorithm performed well in a pedestrian scenario increasing cell capacity to double for video stream with ‘Rapid Movement’ (RM) content. For ‘Medium Movement’ (MM) video content, the capacity was increased about 20% compared to MLWDF and by 40% compared to EXP-PF. In a vehicular scenario, the proposed scheme managed to enhance the cell capacity for MM video stream case. The project has led to three publications (IEEE Globecom’12 – QoEMC Workshop, IEEE CCNC’15 and IEEE MMTC E-letter/May-2015). A journal paper is in preparation.

The game industry is one of the biggest economic sector in the entertainment business whose product rely heavily on the quality of the interactivity to stay relevant. Non-Player Character (NPC) is the main mechanic used for this purpose and it has to be optimised for its designated behaviour. The development process iteratively circulates the results among game designers, game AI developers, and game testers. Automatic optimisation of NPCs to designer’s objective will increase the speed of each iteration, and reduce the overall production time. Previous attempts used entropy evaluation metrics which are difficult to translate the terms to the optimising game and a slight misinterpretation often leads to incorrect measurement. This thesis proposes an alternative method which evaluates generated game data with reference result from the testers. The thesis first presents a reliable way to extract information for NPCs classification called Relative Region Feature (RRF). RRF provides an excellent data compression method, a way to effectively classify, and a way to optimise objective-oriented adaptive NPCs. The formalised optimisation is also proved to work on classifying player skill with the reference hall-of-fame scores. The demonstration are done on the on-line competition version of Ms PacMan. The generated games from participating entries provide challenging optimising problems for various evolutionary optimisers. The thesis developed modified version of CMA-ES and PSO to effectively tackle the problems. It also demonstrates the adaptivity of MCTS NPC which uses the evaluation method. This NPC performs reasonably well given adequate resources and no reference NPC is required.

Introduction: Professional development is an important aspect of working practices for academic librarians. In the past decades the academic library has been subjected to an enormous range of technological and cultural advancements that have constantly required its staff to develop their professional knowledge and understanding. One of the most recent revolutions has been the advent of social media. This advent of a new technology can provide challenges and opportunities but to fully understand these we need to analyse the relationship between social media and various context in more depth. Method: This inductive grounded theory study was iteratively conducted in three different academic libraries in England. At each location an analysis of the social media in use both organisationally and individually was made, and a purposefully selected number of actors was interviewed to gain a deeper understanding of their ideas about social media, CPD and the potential to learn from social media. Findings: The findings demonstrate that the managerial view of CPD is not necessarily compatible with the perception, or needs, of practitioners in relation to their professional development. To increase understanding of this domain, the study analysed the underlying factors related to both social media use and perception, and participation in and perception of professional development. This has led to the discovery of a theory that can be summarised as Continuous Professional Adaptation is learning that occurs as a consequence of professional awareness and preparedness through the use of informal networks. Outcome: The grounded theory presented in this document demonstrates a need to expand the dialogue on professional development in academic libraries, in both an academic and practical context, based on the advent of social media. Formal development in the shape of Continuing Professional Development (CPD) is still regarded as the most relevant method of developing human resources due to the prevalence of a managerial approach to professional development. Conclusion: Increasingly academics and practitioners require more than the formal approach to professional development which can be classed as increasingly being irrelevant due to practical limitations of delivery time and mode and speed of changes occurring. This research contributes to that dialogue, providing a unique approach by incorporating social media with professional development.

The Web 2.0 phenomenon has produced a number of technology companies that in various rounds of venture capital funding, have attracted very indicative valuations. Following these rising valuations, Investment Banks took an interest in the sector. However, while the companies concerned seem stable as private entities, their novel approach to business makes their financial characteristics difficult to predict. Parallels are drawn between the 2001 dot-com bubble and the current Web 2.0 sector. This thesis highlights a dependency between modern highlights a dependency between modern web companies, and the established technology sector. It aims to identify the extent to which the contemporary technology sector (encompassing Web 2.0) has exhibited characteristics similar to those of the dot-com bubble. to that end, this thesis identifies characteristics of modern and historic bubbles, and uses them to formulate a hypothetical set of indicators, in the form of a conceptual model. To determine whether these indicators exist in real data, a novel, repeatable statistical test is developed. It first identifies statistical heuristics representative of bubble circumstances, and then compares other periods to them. Thus, given sufficient data, any period may be tested. Periods are analysed prior, during and after the dot-com bubble. During the dot-com bubble, consistently strong venture capital activity is observed, and linked to the growth in people using the internet. This is indicative of the poor decision-making by investors, documented at the time. In recent periods, patterns in venture capital investment describe an industry that is much more cautious that before, reducing the probability of the formation of a similar bubble. Looking at the past, this thesis observes investor activity that 'caused' the dot-com bubble as early as 1995-96, which raises questions about when the bubble started, and the lead-times on market collapses.

This thesis presents the case for adopting comics in the design, visualisation and evaluation of interactive stories. The potential for comics to be employed in the representation of interactive story-driven material has been identified in previous work. However, there is a lack of theory or evidence upon which an informed approach can be based. Consequently, this thesis contributes a process for employing branching comics to design and visualise interactive stories informed by previous approaches to stories, interactive stories and visualisations. It is argued that comics have several advantages over previous methods of designing interactive stories due to their inherent structural compatibility with visualising hierarchies of abstraction of story content. A series of studies are conducted to demonstrate how comics can be employed to visualise abstraction levels, and how branching comics can be employed to evaluate interactive stories. Qualitative and quantitative methods related to both user experience and comprehension are employed, which demonstrate the advantages in the use of comics to explore a range of different phenomena related to creating, interpreting and using interactive stories.

The recently released High Efficiency Video Coding (HEVC) standard, which halves the transmission bandwidth requirement of encoded video for almost the same quality when compared to H.264/AVC, and the availability of increased network bandwidth (e.g. from 2 Mbps for 3G networks to almost 100 Mbps for 4G/LTE) have led to the proliferation of video streaming services. Based on these major innovations, the prevalence and diversity of video application are set to increase over the coming years. However, the popularity and success of current and future video applications will depend on the perceived quality of experience (QoE) of end users. How to measure or predict the QoE of delivered services becomes an important and inevitable task for both service and network providers. Video quality can be measured either subjectively or objectively. Subjective quality measurement is the most reliable method of determining the quality of multimedia applications because of its direct link to users’ experience. However, this approach is time consuming and expensive and hence the need for an objective method that can produce results that are comparable with those of subjective testing. In general, video quality is impacted by impairments caused by the encoder and the transmission network. However, videos encoded and transmitted over an error-prone network have different quality measurements even under the same encoder setting and network quality of service (NQoS). This indicates that, in addition to encoder settings and network impairment, there may be other key parameters that impact video quality. In this project, it is hypothesised that video content type is one of the key parameters that may impact the quality of streamed videos. Based on this assertion, parameters related to video content type are extracted and used to develop a single metric that quantifies the content type of different video sequences. The proposed content type metric is then used together with encoding parameter settings and NQoS to develop content-based video quality models that estimate the quality of different video sequences delivered over IP-based network. This project led to the following main contributions: (1) A new metric for quantifying video content type based on the spatiotemporal features extracted from the encoded bitstream. (2) The development of novel subjective test approach for video streaming services. (3) New content-based video quality prediction models for predicting the QoE of video sequences delivered over IP-based networks. The models have been evaluated using subjective and objective methods.

Compared to information systems that work the same for all users and contexts, systems that utilise contextual information have greater potential to help a user identify relevant information more quickly and more accurately. Contextual information comes in a variety of flavours, often derived from records of past interactions between a user and the information system. It can be individual- or group-based. The motivation for our work is as follows. First, instead of looking at Web searching or browsing, which has been studied extensively, we focus our attention on Web sites. Such collections can be notoriously difficult to search or explore. If we could learn from past user interactions what information needs can be satisfied by which -documents, we would be in a position to help a new user to get to the required information much more rapidly. Hence, we harness the search behaviour of cohorts of users instead of individual users, turning it automatically into a profile which can then be used to assist other users of the same cohort. Finally, we are interested in exploring how such a profile is best utilised for profile-based summarisation of the collection at hand in a navigation scenario in which such summaries can be displayed as hover text as a user moves the mouse over a link. The process of acquiring the profile is not a research interest here; we simply adopt a biologically inspired method that resembles the idea of ant colony optimisation (AGO). This has been shown to work well in a variety of application areas. The model can be built in a continuous learning cycle that exploits search patterns as recorded in typical query log files. The main focus of this thesis will be on using the model in profile-based summarisation to generate summaries of documents for navigation support. Our research explores different single-document and multi- document summarisation techniques, some of which use the profile and some of which do not. We perform task-based evaluations of these different techniques - and hence of the impact of the profile and profile-based summarisation - in the context of Web site navigation. The experimental results demonstrate that profile-based summarisation to assist users in navigation tasks can significantly outperform generic summarisation as well as a standard Web site without such assistance.

Twitter is a microblogging application used for sending and retrieving instant on-line messages of not more than 140 characters. There has been a surge in Twitter activities since its launch in 2006 as well as steady increase in event detection research on Twitter data (tweets) in recent years. With 284 million monthly active users Twitter has continued to grow both in size and activity. The network is rapidly changing the way global audience source for information and influence the process of journalism [Newman, 2009]. Twitter is now perceived as an information network in addition to being a social network. This explains why traditional news media follow activities on Twitter to enhance their news reports and news updates. Knowing the significance of the network as an information dissemination platform, news media subscribe to Twitter accounts where they post their news headlines and include the link to their on-line news where the full story may be found. Twitter users in some cases, post breaking news on the network before such news are published by traditional news media. This can be ascribed to Twitter subscribers' nearness to location of events. The use of Twitter as a network for information dissemination as well as for opinion expression by different entities is now common. This has also brought with it the issue of computational challenges of extracting newsworthy contents from Twitter noisy data. Considering the enormous volume of data Twitter generates, users append the hashtag (#) symbol as prefix to keywords in tweets. Hashtag labels describe the content of tweets. The use of hashtags also makes it easy to search for and read tweets of interest. The volume of Twitter streaming data makes it imperative to derive Topic Detection and Tracking methods to extract newsworthy topics from tweets. Since hashtags describe and enhance the readability of tweets, this research is developed to show how the appropriate use of hashtags keywords in tweets can demonstrate temporal evolvements of related topic in real-life and consequently enhance Topic Detection and Tracking on Twitter network. We chose to apply our method on Twitter network because of the restricted number of characters per message and for being a network that allows sharing data publicly. More importantly, our choice was based on the fact that hashtags are an inherent component of Twitter. To this end, the aim of this research is to develop, implement and validate a new approach that extracts newsworthy topics from tweets' hashtags of real-life topics over a specified period using Association Rule Mining. We termed our novel methodology Transaction-based Rule Change Mining (TRCM). TRCM is a system built on top of the Apriori method of Association Rule Mining to extract patterns of Association Rules changes in tweets hashtag keywords at different periods of time and to map the extracted keywords to related real-life topic or scenario. To the best of our knowledge, the adoption of dynamics of Association Rules of hashtag co-occurrences has not been explored as a Topic Detection and Tracking method on Twitter. The application of Apriori to hashtags present in tweets at two consecutive period t and t + 1 produces two association rulesets, which represents rules evolvement in the context of this research. A change in rules is discovered by matching every rule in ruleset at time t with those in ruleset at time t + 1. The changes are grouped under four identified rules namely 'New' rules, 'Unexpected Consequent' and 'Unexpected Conditional' rules, 'Emerging' rules and 'Dead' rules. The four rules represent different levels of topic real-life evolvements. For example, the emerging rule represents very important occurrence such as breaking news, while unexpected rules represents unexpected twist of event in an on-going topic. The new rule represents dissimilarity in rules in rulesets at time t and t+1. Finally, the dead rule represents topic that is no longer present on the Twitter network. TRCM revealed the dynamics of Association Rules present in tweets and demonstrates the linkage between the different types of rule dynamics to targeted real-life topics/events. In this research, we conducted experimental studies on tweets from different domains such as sports and politics to test the performance effectiveness of our method. We validated our method, TRCM with carefully chosen ground truth. The outcome of our research experiments include: Identification of 4 rule dynamics in tweets' hashtags namely: New rules, Emerging rules, Unexpected rules and 'Dead' rules using Association Rule Mining. These rules signify how news and events evolved in real-life scenario. Identification of rule evolvements on Twitter network using Rule Trend Analysis and Rule Trace. Detection and tracking of topic evolvements on Twitter using Transaction-based Rule Change Mining TRCM. Identification of how the peculiar features of each TRCM rules affect their performance effectiveness on real datasets.

In social networks, analysing the explicit interactions among users can help in inferring hierarchical relationships and roles that may be implicit. In this thesis, we focus on two objectives: detecting hierarchical relationships between users and inferring the hierarchical roles of users interacting via the same online communication medium. In both cases, we show that considering the temporal dimension of interaction substantially improves the detection of relationships and roles. The first focus of this thesis is on the problem of inferring implicit relationships from interactions between users. Based on promising results obtained by standard link-analysis methods such as PageRank and Rooted-PageRank (RPR), we introduce three novel time-based approaches, \Time-F" based on a defined time function, Filter and Refine (FiRe) which is a hybrid approach based on RPR and Time-F, and Time-sensitive Rooted-PageRank (T-RPR) which applies RPR in a way that takes into account the time-dimension of interactions in the process of detecting hierarchical ties. We experiment on two datasets, the Enron email dataset to infer managersubordinate relationships from email exchanges, and a scientific publication coauthorship dataset to detect PhD advisor-advisee relationships from paper co-authorships. Our experiments demonstrate that time-based methods perform better in terms of recall. In particular T-RPR turns out to be superior over most recent competitor methods as well as all other approaches we propose. The second focus of this thesis is examining the online communication behaviour of users working on the same activity in order to identify the different hierarchical roles played by the users. We propose two approaches. In the first approach, supervised learning is used to train different classification algorithms. In the second approach, we address the problem as a sequence classification problem. A novel sequence classification framework is defined that generates time-dependent features based on frequent patterns at multiple levels of time granularity. Our framework is a exible technique for sequence classification to be applied in different domains. We experiment on an educational dataset collected from an asynchronous communication tool used by students to accomplish an underlying group project. Our experimental findings show that the first supervised approach achieves the best mapping of students to their roles when the individual attributes of the students, information about the reply relationships among them as well as quantitative time-based features are considered. Similarly, our multi-granularity pattern-based framework shows competitive performance in detecting the students' roles. Both approaches are significantly better than the baselines considered.

Data compression is used in our everyday life to improve computer interaction or simply for storage purposes. Lossless data compression refers to those techniques that are able to compress a file in such ways that the decompressed format is the replica of the original. These techniques, which differ from the lossy data compression, are necessary and heavily used in order to reduce resource usage and improve storage and transmission speeds. Prior research led to huge improvements in compression performance and efficiency for general purpose tools which are mainly based on statistical and dictionary encoding techniques. Extensible Markup Language (XML) is based on redundant data which is parsed as normal text by general-purpose compressors. Several tools for compressing XML data have been developed, resulting in improvements for compression size and speed using different compression techniques. These tools are mostly based on algorithms that rely on variable length encoding. XML Schema is a language used to define the structure and data types of an XML document. As a result of this, it provides XML compression tools additional information that can be used to improve compression efficiency. In addition, XML Schema is also used for validating XML data. For document compression there is a need to generate the schema dynamically for each XML file. This solution can be applied to improve the efficiency of XML compressors. This research investigates a dynamic approach to compress XML data using a hybrid compression tool. This model allows the compression of XML data using variable and fixed length encoding techniques when their best use cases are triggered. The aim of this research is to investigate the use of fixed length encoding techniques to support general-purpose XML compressors. The results demonstrate the possibility of improving on compression size when a fixed length encoder is used to compress most XML data types.

Can a game help you write a meaningful story with others? Academic discussion of collaborative story-writing games usually contains reference to Surrealist game Exquisite Corpse, known to many as Consequences. In it, a game rule prohibits players, writing in turn, from reading most of the story written before their contribution. This rule promotes a fragmented form of narrative which, although often inventive and entertaining, does not often fulfil the normal requirements of a story. Is it possible to design a writing game with different rules that instead promote the production of a cohesive and meaningful story? In order to explore the possibilities of game-based story writing, the researcher developed two web-based games that formed the online platform Storyjacker (www.storyjacker.net). These were produced via an iterative design methodology which involved cyclical phases of software development and user testing. Design was also informed by a multi-disciplinary literature review and analysis of four other online writing platforms. Following the design phase, a selection of the stories that had been produced during Storyjacker testing were then rated and commented on by an expert reading panel, made up of creative writing academics and literary industry professionals. The panel’s ratings and comments informed a final analysis of the Storyjacker games as methods for creating meaningful stories. The research found that bespoke creative writing games do produce relatively meaningful stories. Plotlines in emerging stories motivate and entertain players through a joint sense of purpose. Writing games are also effective as a collaborative framework because they allow participants to work creatively together without feeling vulnerable in front of other players. However, the research finds that there are limits to the meaningfulness of stories written through games. By aligning gameplay with linear plot development, a feature of both Storyjacker games, there is no opportunity or motivation for players to go back and redraft stories. This, ultimately, has the most negative effect on story meaningfulness as this is the normal way that writers make sense of their stories. Entertaining game features, such as turn-based writing challenges, can also negatively impact story quality by distracting the writer from the overall storytelling task. Complicating the debate on meaningfulness, the thesis also finds that texts written by games typically contain two discernable narratives: the story narrative and a description of the event of playing the game. This both enriches the text and makes it more difficult readers to decode as a meaningful story. Further practice-led research is needed to explore ways in which story games could incorporate a drafting process. This could significantly improve the meaningfulness of stories produced in this way. In the broader context of design, successful features in the Storyjacker games should be trialled in other digital interfaces to see if they help users perform other creative or subjective tasks. Finally, more research should be conducted on the effects of these collaborative games in broader educational, creative and organisational contexts.

This thesis analyses contemporary constructions of "pastness", as nostalgically expressed by an online community on the popular microblogging platform Tumblr. Drawing on original ethnographic and semiotic work, as well as on a wide range of sources centred on its medical past and subsequent literary, psychological and social acculturation throughout the 19th and 20th centuries, I plead for an understanding of nostalgia not as the reactionary, inauthentic, naive or irretrievably commercial social disease that (similarly nostalgic) mainstream academic discourses have framed it as, but rather as a pervasive, protean and often misunderstood complication, or consequence of modernity — one that, I continue to argue, can facilitate culturally complex, if ambivalent, forms of personal expression. In the specific case of my participants, young people articulating their vision of themselves and of the world through publically available (yet also privately informed and interpersonally mediated), digitally synesthetic narratives, I show how the emotion becomes further entangled in biographical, transbiographical, historical and transhistorical, as well as aesthetic, emotional and artistic contingencies. This amounts to a highly idiosyncratic, symbolically coherent reflexive ideology, the cultural significance of which my thesis will contextualize, interpret and draw important conclusions from.

Graph data is becoming more and more pervasive. Indeed, services such as Social Networks or the Semantic Web can no longer rely on the traditional relational model, as its structure is somewhat too rigid for the applications they have in mind. For this reason we have seen a continuous shift towards more non-standard models. First it was the semi-structured data in the 1990s and XML in 2000s, but even such models seem to be too restrictive for new applications that require navigational properties naturally modelled by graphs. Social networks fit into the graph model by their very design: users are nodes and their connections are specified by graph edges. The W3C committee, on the other hand, describes RDF, the model underlying the Semantic Web, by using graphs. The situation is quite similar with crime detection networks and tracking workflow provenance, namely they all have graphs inbuilt into their definition. With pervasiveness of graph data the important question of querying and maintaining it has emerged as one of the main priorities, both in theoretical and applied sense. Currently there seem to be two approaches to handling such data. On the one hand, to extract the actual data, practitioners use traditional relational languages that completely disregard various navigational patterns connecting the data. What makes this data interesting in modern applications, however, is precisely its ability to compactly represent intricate topological properties that envelop the data. To overcome this issue several languages that allow querying graph topology have been proposed and extensively studied. The problem with these languages is that they concentrate on navigation only, thus disregarding the data that is actually stored in the database. What we propose in this thesis is the ability to do both. Namely, we will study how query languages can be designed to allow specifying not only how the data is connected, but also how data changes along paths and patterns connecting it. To this end we will develop several query languages and show how adding different data manipulation capabilities and different navigational features affects the complexity of main reasoning tasks. The story here is somewhat similar to the early success of the relational data model, where theoretical considerations led to a better understanding of what makes certain tasks more challenging than others. Here we aim for languages that are both efficient and capable of expressing a wide variety of queries of interest to several groups of practitioners. To do so we will analyse how different requirements affect the language at hand and at the end provide a good base of primitives whose inclusion into a language should be considered, based on the applications one has in mind. Namely, we consider how adding a specific operation, mechanism, or capability to the language affects practical tasks that such an addition plans to tackle. In the end we arrive at several languages, all of them with their pros and cons, giving us a good overview of how specific capabilities of the language affect the design goals, thus providing a sound basis for practitioners to choose from, based on their requirements.

XML (Extensible Markup Language) is a language used in data representation and storage, and transmission and manipulation of data. Excessive memory consumption is an important challenge when representing XML documents in main memory. Document Object Model (DOM) APIs are used in a processing level that provides access to all parts of XML documents through the navigation operations. Although DOM serves as a a general purpose tool that can be used in different applications, it has high memory cost particularly if using naïve. The space usage of DOM has been reduced significantly while keeping fast processing speeds, by use of succinct data structures in SiXDOM [1]. However, SiXDOM does not explore in depth XML data compression principles to improve in-memory space usage. Such XML data compression techniques have been proven to be very effective in on-disk compression of XML document. In this thesis we propose a new approach to represent XML documents in-memory using XML data compression ideas to further reduce space usage while rapidly supporting operations of the kind supported by DOM. Our approach is based upon a compression method [2] which represents an XML document as a directed acyclic graph (DAG) by sharing common subtrees. However, this approach does not permit the representation of attributes and textual data, and furthermore, a naive implementation of this idea gives very poor space usage relative to other space-efficient DOM implementations [1]. In order to realise the potential of this compression method as an in-memory representation, a number of optimisations are made by application of succinct data structures and variablelength encoding. Furthermore, a framework for supporting attribute and textual data nodes is introduced. Finally, we propose a novel approach to representing the textual data using Minimal Perfect Hashing(MPH). We have implemented our ideas in a software library called DAGDOMand performed extensive experimental evaluation on a number of standard XML files. DAGDOM yields a good result and we are able to obtain significant space reductions over existing space-efficient DOM implementations (typically 2 to 5 times space reduction), with very modest degradations in CPU time for navigational operations.

The main hypothesis of the thesis is that between two systems identical in functionality and usability, differences in aesthetics may positively influence users perceived usability. To date, a narrow focus on the engineering aspects of aesthetics has adversely affected the scope and success of experiments, therefore previous work in the field needed to be revisited. The thesis reviews literature and theory in usability and aesthetics, the latter from the point of the view of philosophy, theory, and application. It also explores the relationship between aesthetics, usability and user engagement; discusses a distinct new trend research that identifies a link between beauty and perceived usability of website interaction; and develops a pilot for an experimental methodology. Based on conclusions from the review of the field of usability, two experiments where designed and carried out, an independent measures and repeated measures. The findings of these experiments confirmed the hypothesis that perceived usability was positively influenced by higher aesthetics.