Search

In modern days, making recommendation for news articles poses a great challenge due to vast amount of online information. However, providing personalized recommendations from news articles, which are the sources of condense textual information is not a trivial task. A recommendation system needs to understand both the textual information of a news article, and the user contexts in terms of long-term and temporary preferences via the user’s historic records. Unfortunately, many existing methods do not possess the capability to meet such need. In this work, we propose a neural deep news recommendation model called CupMar, that not only is able to learn the user-profile representation in different contexts, but also is able to leverage the multi-aspects properties of a news article to provide accurate, personalized news recommendations to users. The main components of our CupMar approach include the News Encoder and the User-Profile Encoder. Specifically, the News Encoder uses multiple properties such as news category, knowledge entity, title and body content with advanced neural network layers to derive informative news representation, while the User-Profile Encoder looks through a user’s browsed news, infers both of her long-term and recent preference contexts to encode a user representation, and finds the most relevant candidate news for her. We evaluate our CupMar model with extensive experiments on the popular Microsoft News Dataset (MIND), and demonstrate the strong performance of our approach.

Session-based recommender systems (SBRSs) aim at predicting the next item via learning the dynamic and short-term preferences of users. Most of the existing SBRSs usually make predictions based on the intra-session dependencies embedded in session information only, ignoring more complex inter-session dependencies and other available side information (e.g., item attributes, users), which in turn greatly limits the improvement of the recommendation accuracy. In order to effectively extract both intra- and inter-session dependencies from not only the session information but also the side information, to further improve the accuracy of next-item recommendations, we propose a novel hypergraph learning (HL) framework. The HL framework mainly contains three modules, i.e., a hypergraph construction module, a hypergraph learning module, and a next-item prediction module. The hypergraph construction module constructs a hypergraph to connect the users, items and item attributes together in a unified way. Then, the hypergraph learning module learns the informative latent representation for each item by extracting both intra- and inter-session dependencies embedded in the constructed hypergraph. Also, a latent representation for each user is learned. After that, the learned latent representations are fed into the next-item prediction module for next-item recommendations. We conduct extensive experiments on two real-world datasets. The experimental results show that our HL framework outperforms the state-of-the-art approaches.

Twitter, as one of the largest Online Social Network (OSN) platforms, is a popular media for online social communication and information dissemination. The trustworthy evaluation of information and people becomes crucial for maintaining an open and healthy OSN for our society. In this work, we develop a Coupled Dual Networks Trust Ranking (CoRank) method to evaluate the trustworthiness of users and tweets by analysing user/tweet behaviours on Twitter. We propose a model to capture the complex characteristics and relations of both users and tweets and calculate their trust values. Our approach goes beyond the existing solutions that use a single network to link both users and tweets. A set of experiments have been conducted against the real data collected from Twitter. The experimental results show the effectiveness, robustness, and time complexity of the proposed method. We also compare our solution with three baseline methods PageRank, TURank, and Weighted PageRank, to show how our approach outperforms the existing ones.

This paper studies a link-text algorithm to model scientific documents by citation influences, which is applied to document clustering and influence prediction. Most existing link-text algorithms ignore the different weights of citation influences that cited documents have on the corresponding citing document. In fact, citation influences reveal the latent structure of citation networks which is more accurate to describe the knowledge flow than the original citation structure. In this study, a citation influence is modeled as a weight of linear combination that approximates the text of a document by the content of its citations. Then, we present a novel matrix factorization algorithm, called Citation-Influences-Text Nonnegative Matrix Factorization (CIT-NMF), which incorporates text and citations to obtain better document representations by learning influence weights. In addition, an efficient optimization method is derived to solve the optimization problem. Experimental results on several real datasets show satisfactory improvements over the baseline models.

Illness severity prediction (ISP) is crucial for caregivers in the intensive care unit (ICU) while saving the life of patients. Existing ISP methods fail to provide sufficient evidence for the time-critical decision making in the dynamic changing environment. Moreover, the correlated temporal features in multivariate time-series are rarely be considered in existing machine learning-based ISP models. Therefore, in this paper, we propose a novel interpretable analysis framework which simultaneously analyses organ systems differentiated based on the pathological and physiological evidence to predict illness severity of patients in ICU. It not only timely but also intuitively reflects the critical conditions of patients for caregivers. In particular, we develop a deep interpretable learning model, namely AMRNN, which is based on the Multi-task RNNs and Attention Mechanism. Physiological features of each organ system in multivariate time series are learned by a single Long-Short Term Memory unit as a dedicated task. To utilize the functional and temporal relationships among organ systems, we use a shared LSTM task to exploit correlations between different learning tasks for further performance improvement. Real-world clinical datasets (MIMIC-III) are used for conducting extensive experiments, and our method is compared with the existing state-of-the-art methods. The experimental results demonstrated that our proposed approach outperforms those methods and suggests a promising way of evidence-based decision support.

Text categorization is widely characterized as a multi-label classification problem. Robust modeling of the semantic similarity between a query text and training texts is essential to construct an effective and accurate classifier. In this paper, we systematically investigate the Web page/text classification problem via integrating sparse representation with random measurements. In particular, we first adopt a very sparse data-independent random measurement matrix to map the original high dimensional text feature space to a lower dimensional space without loss of key information. We then propose a generic sparse representation method to obtain the sparse solution by decoding the semantic correlations between the query text and entire training samples. Based on the above method, we also design and examine a series of rules by taking advantage of the sparse coefficients to propagate multiple labels for the given query texts. We have conducted extensive experiments using real-world datasets to examine our proposed approach, and the results show the effectiveness of the proposed approach.

Twitter has recently emerged as a popular microblogging service that has 284 million monthly active users around the world. A part of the 500 million tweets posted on Twitter everyday are personal observations of immediate environment. If provided with time and location information, these observations can be seen as sensory readings for monitoring and localizing objects and events of interests. Location information on Twitter, however, is scarce, with less than 1% of tweets have associated GPS coordinates. Current researches on Twitter location inference mostly focus on city-level or coarser inference, and cannot provide accurate results for fine-grained locations. We propose an event monitoring system for Twitter that emphasizes local events, called SNAF (Sense and Focus). The system filters personal observations posted on Twitter and infers location of each report. Our extensive experiments with real Twitter data show that, the proposed observation filtering approach can have about 22% improvement over existing filtering techniques, and our location inference approach can increase the location accuracy by up to 36% within the 3km error range. By aggregating the observation reports with location information, our prototype event monitoring system can detect real world events, in many case earlier than news reports.

Automatic annotation is an essential technique for effectively handling and organizing Web objects (e.g., Web pages), which have experienced an unprecedented growth over the last few years. Automatic annotation is usually formulated as a multi-label classification problem. Unfortunately, labeled data are often time-consuming and expensive to obtain. Web data also accommodate much richer feature space. This calls for new semi-supervised approaches that are less demanding on labeled data to be effective in classification. In this paper, we propose a graph-based semi-supervised learning approach that leverages random walks and l1 sparse reconstruction on a mixed object-label graph with both attribute and structure information for effective multi-label classification. The mixed graph contains an object-affinity subgraph, a label-correlation subgraph, and object-label edges with adaptive weight assignments indicating the assignment relationships. The object-affinity subgraph is constructed using l1 sparse graph reconstruction with extracted structural meta-text, while the label-correlation subgraph captures pairwise correlations among labels via linear combination of their co-occurrence similarity and kernel-based similarity. A random walk with adaptive weight assignment is then performed on the constructed mixed graph to infer probabilistic assignment relationships between labels and objects. Extensive experiments on real Yahoo! Web datasets demonstrate the effectiveness of our approach.

© 2015, Springer Science+Business Media New York. Recently, newly invented features (e.g. Fisher vector, VLAD) have achieved state-of-the-art performance in large-scale video analysis systems that aims to understand the contents in videos, such as concept recognition and event detection. However, these features are in high-dimensional representations, which remarkably increases computation costs and correspondingly deteriorates the performance of subsequent learning tasks. Notably, the situation becomes even worse when dealing with large-scale video data where the number of class labels are limited. To address this problem, we propose a novel algorithm to compactly represent huge amounts of unconstrained video data. Specifically, redundant feature dimensions are removed by using our proposed feature selection algorithm. Considering unlabeled videos that are easy to obtain on the web, we apply this feature selection algorithm in a semi-supervised framework coping with a shortage of class information. Different from most of the existing semi-supervised feature selection algorithms, our proposed algorithm does not rely on manifold approximation, i.e. graph Laplacian, which is quite expensive for a large number of data. Thus, it is possible to apply the proposed algorithm to a real large-scale video analysis system. Besides, due to the difficulty of solving the non-smooth objective function, we develop an efficient iterative approach to seeking the global optimum. Extensive experiments are conducted on several real-world video datasets, including KTH, CCV, and HMDB. The experimental results have demonstrated the effectiveness of the proposed algorithm.

The ability to track and trace individual items, especially through large-scale and distributed networks, is the key to realizing many important business applications such as supply chain management, asset tracking, and counterfeit detection. Networked RFID (radio frequency identification), which uses the Internet to connect otherwise isolated RFID systems and software, is an emerging technology to support traceability applications. Despite its promising benefits, there remain many challenges to be overcome before these benefits can be realized. One significant challenge centers around dealing with uncertainty of raw RFID data. In this paper, we propose a novel framework to effectively manage the uncertainty of RFID data in large scale traceability networks. The framework consists of a global object tracking model and a local RFID data cleaning model. In particular, we propose a Markov-based model for tracking objects globally and a particle filter based approach for processing noisy, low-level RFID data locally. Our implementation validates the proposed approach and the experimental results show its effectiveness.

Advances in the Web and mobile technologies offer new and exciting services to users in various domains such as e-business, healthcare, entertainment, and scientific activities. With the advances in mobile technologies, the Web gets richer in its contents and usability. In addition to its widespread usability in the e-commerce, education, and media, the Web has established its roots in exchanging and sharing information in social communities. For example, user-generated contents in the social networks have transformed the Web into a new era of the Web 2.0—these contents comprise a wealth of information often involving multimedia data such as audio, video, pictures and blogs. Indeed, the Web is slowly evolving from current “Web of Content”, “Web of People”, into “Web of Things and Services”, given the recent advances such as radio-frequency identification (RFID), sensor networks, and service-oriented computing (SOC). These new developments however introduce many new challenging and unsolved problems. One of the major issues is to deal with the resource scarcity in mobile environments. Though mobile devices and wireless networks are improving their resources, they are still far behind the wired networks and desktop PCs in terms of bandwidth, memory, processing speed and battery power. In addition, users expect mobile services to be intelligent and context-aware.