Search

The maximum likelihood estimation is a probabilistic inferencing model of community connectivity in large networks. In general, only the adjacency matrix is utilized to perform community structure parameter inference. Although there are recent examples that combine connectivity and attribute information for community detection, our model is an enhanced overlapping community detection model that combines adjacency spectral embedding with maximum likelihood estimation. This provides the flexibility of complex networks to increase connectivity information through measurements from attribute embedding. The attribute information can be effectively captured and transformed by attribute embedding to encode the combination with structure information. Then, the link strength among communities is designed to adjust the impact of these structural information on community generation based on the contribution of the structure to the clusters, and the node assignment allow for the nature of the real network (overlapping and outliers). Experiments highlight attributed networks in which attributed community detection task provides satisfactory performance.

With the rapid development of information technology, the amount of textual data generated in biomedical field becomes larger and larger. Biomedical event extraction, which is a fundamental information extraction task, has gained a growing interest in biomedical community. Although researchers have proposed various approaches to this task, the performance is still undesirable since previous approaches fail to model biomedical documents effectively. In this paper, we propose an end-to-end framework for document-level joint biomedical event extraction. To better capture the complex relationships among contexts in biomedical documents, a two-level modeling approach is introduced for biomedical documents. More specifically, the dependency-based GCN and hypergraph are used to model local context and global context in each biomedical document, respectively. In addition, a fine-grained interaction mechanism is proposed to model effectively the interaction between local and global contexts to learn better contextualized representations for biomedical event extraction. Comprehensive experiments on two widely used datasets are conducted and the results demonstrate the effectiveness of the proposed framework over state-of-the-art methods.

To better learn the dependency relationship between nodes, we address the relationship extraction task by capturing rich contextual dependencies based on the attention mechanism, and using distributional reinforcement learning to generate optimal relation information representation. This method is called Dual Attention Graph Convolutional Network (DAGCN), to adaptively integrate local features with their global dependencies. Specifically, we append two types of attention modules on top of GCN, which model the semantic interdependencies in spatial and relational dimensions, respectively. The position attention module selectively aggregates the feature at each position by a weighted sum of the features at all positions of nodes internal features. Meanwhile, the relation attention module selectively emphasizes interdependent node relations by integrating associated features among all nodes. We sum the outputs of the two attention modules and use reinforcement learning to predict the classification of nodes relationship to further improve feature representation which contributes to more precise extraction results. The results on the TACRED and SemEval datasets show that the model can obtain more useful information for relational extraction tasks, and achieve better performances on various evaluation indexes.

With the rapid development of information technology, information is generated at an unprecedented rate. Users are in great need of recommender systems to provide the potential friends or interested items for them. Social (i.e. friend) recommendation and behavior (i.e. item) recommendation are two types of popular services in real-world applications. Although researchers have proposed various models for each task, a unified model to address both tasks elegantly and effectively is still in demand. In this paper, we propose a model called SBRNE which integrates social and behavior recommendations into a unified framework through modeling social and behavior information simultaneously. Specifically, SBRNE models social and behavior information simultaneously via employing users’ latent interests as a bridge, and derives improved performance on both social and behavior recommendation tasks. In addition, by introducing an efficient network embedding procedure, users’ latent representations are advanced, and effectiveness and efficiency of recommendation tasks are improved accordingly. Results on both real-world and synthetic datasets demonstrate that: 1). SBRNE outperforms selected baselines on social and behavior recommendation tasks; 2). SBRNE performs stable on recommendation tasks for cold-start users; 3). The network embedding procedure can improve the effectiveness of SBRNE; 4). The hyper-parameter learning procedure can improve both the effectiveness and efficiency of SBRNE.

A good document summary should summarize the core content of the text. Research on automatic text summarization attempts to solve this problem. The encoder-decoder model is widely used in text summarization research. Soft attention is used to obtain the required contextual semantic information during decoding. However, due to the lack of access to the key features, the generated summary deviates from the core content. In this paper, we proposed an encoder-decoder model based on a double attention pointer network (DAPT). In DAPT, the self-attention mechanism collects key information from the encoder, the soft attention and the pointer network generate more coherent core content, and the fusion of both generates accurate and coherent summaries. In addition, the improved coverage mechanism is used to address the repetition problem and improve the quality of the generated summaries. Simultaneously, scheduled sampling and reinforcement learning (RL) are combined to generate new training methods to optimize the model. Experiments on the CNN/Daily Mail dataset and the LCSTS dataset show that our model performs as well as many state-of-the-art models. The experimental analysis shows that our model achieves higher summarization performance and reduces the occurrence of repetition.

Community search has been wildly studied to retrieve relevant cluster from attributed graph for a given set of sample nodes. However, existing well-studied methods aim at searching a community containing sample nodes, and fail to capture communities without sample nodes but are similar with user preference deduced from the given sample nodes. To this end, we propose a community search method that is capable of finding multi-communities with user's preference using few given sample nodes and simultaneously identify outliers in attributed network. The method is termed as Integrating user Preference into Multi-community and Outliers Search (IPMOS), which collaborates user's preference into the process of searching to find interesting clusters of the entire network. Particularly, the strategy of truncated random walk is first used to expand few sample nodes. And then, the average partition similarity is defined based on exploring strategy to infer attribute subspace as user's latent interest. Finally, multiple communities and outliers in the whole network are detected via fractional-core and structural constraints. Extensive experiments on several synthetic networks and real-world networks with different scales and subjects demonstrate the effectiveness and efficiency of our approach.

The existing attributed community search methods mainly utilize seed nodes to explore a densely connected subgraph which satisfies both structure cohesiveness and attribute homogeneity. However, in many application scenarios, since the diversity of attribute types and the uncertainty of seed node quality, it is challenging to develop stable community search methods. To solve these problems, we propose a two-stage community search method based on Seed Replacement and joint Random Walk (SRRW). First, we use the overlapping clustering method to cluster nodes. Second, we propose a dynamic local clustering coefficient and cluster center membership coefficient to evaluate the quality of nodes. Third, we propose a seed replacement strategy to replace the seed node with the core member of the target community. Finally, we use joint random walk and parallel conductance value to detect a local community. Experimental results on synthetic and realworld networks show that SRRW is more robust to the seed-dependent problem and the seed-invalid problem than state-of-the-art algorithms.

Traditional recommendation algorithms predict the latent interest of an active user by collecting rating information from other similar users or items. Recently, more and more recommendation systems attempt to involve social relations to improve recommendation performance. However, the existing works either leave out the user reliability or cannot capture the correlation between two users who are similar but not socially connected. Besides, they also take the trust value between users either 0 or 1, thus degenerating the prediction accuracy. In this paper, we propose an efficient social affect model, multi-affect(ed), for recommendation via incorporating both users’ reliability and influence propagation. Specifically, the model contains two main components, i.e., computation of user reliability and influence propagation, designing of user-shared feature space. Firstly, a reliability calculation strategy based on user similarity is developed for measuring the recommendation accuracy between users. Then, the factor of influence propagation relationship among users is taken into consideration. Finally, the multi-affect(ed) model is developed with user-shared feature space to generate the predicted ratings. Experimental results demonstrate that the proposed model achieves better accuracy than other counterparts recommendation techniques.

Community detection in the network has become an invaluable tool to explore and reveal the internal organization of nodes. In particular, the target community detection focuses on discovering the “local” links within and connecting to the target community related to user's preference, which refers to a limited number of nodes in the whole network. A few works in the literature discuss the target community detection. In this paper, we propose a target community detection with user's preference and attribute subspace. Our method utilizes not only network structure but also node attributes within a certain subspace to quantify both internal consistency and external separability, which is able to capture a user preferred target community. First, the similarity between nodes is calculated via both attributes and structures, and the center node set of the target community can be obtained by extending the sample node given by the user with its neighbors. Second, an attribute subspace calculation method with entropy weights is established based on the center node set, and the attribute subspace of the target community can thus be deduced. Finally, the target community quality, which is the combination of internal connectivity and external separability, is defined, based on which the target community with a user's preference can be detected. The experimental results on both synthetic network and real-world network datasets demonstrated the efficiency and effectiveness of the proposed algorithm.

Image captioning methods with attention mechanism are leading this field, especially models with global and local attention. But there are few conventional models to integrate the relationship information between various regions of the image. In this paper, this kind of relationship features are embedded into the fused attention mechanism to explore the internal visual and semantic relations between different object regions. Besides, to alleviate the exposure bias problem and make the training process more efficient, we combine Generative Adversarial Network with Reinforcement Learning and employ the greedy decoding method to generate a dynamic baseline reward for self-critical training. Finally, experiments on MSCOCO datasets show that the model can generate more accurate and vivid image captioning sentences and perform better in multiple prevailing metrics than the previous advanced models.

Knowledge tracing (KT) lies at the core of intelligent education, which aims to diagnose students’ changing knowledge level over time based on their historical performance. Most of the existing KT models either ignore the significance of Q-matrix associated exercises with knowledge concepts (KCs) or fail to eliminate the subjective tendency of experts within the Q-matrix, thus it is insufficient for capturing complex interaction between students and exercises. In this paper, we propose a novel Relevance-Aware Q-matrix Calibration method for knowledge tracing (RAQC), which incorporates the calibrated Q-matrix into Long Short-Term Memory (LSTM) network to model the complex students’ learning process, for getting both accurate and interpretable diagnosis results. Specifically, we first leverage the message passing mechanism in Graph Convolution Network (GCN) to fully exploit the high-order connectivity between exercises and KCs for obtaining a potential KC list. Then, we propose a Q-matrix calibration method by using relevance scores between exercises and KCs to mitigate the problem of subjective bias existed in human-labeled Q-matrix. After that, the embedding of each exercise aggregated the calibrated Q-matrix with the corresponding response log is fed into the LSTM to tracing students’ knowledge states (KS). Extensive experimental results on two real-world datasets show the effectiveness of the proposed method.

Automatically generating a human-like description for a given image is a potential research in artificial intelligence, which has attracted a great of attention recently. Most of the existing attention methods explore the mapping relationships between words in sentence and regions in image, such unpredictable matching manner sometimes causes inharmonious alignments that may reduce the quality of generated captions. In this paper, we make our efforts to reason about more accurate and meaningful captions. We first propose word attention to improve the correctness of visual attention when generating sequential descriptions word-by-word. The special word attention emphasizes on word importance when focusing on different regions of the input image, and makes full use of the internal annotation knowledge to assist the calculation of visual attention. Then, in order to reveal those incomprehensible intentions that cannot be expressed straightforwardly by machines, we introduce a new strategy to inject external knowledge extracted from knowledge graph into the encoder-decoder framework to facilitate meaningful captioning. Finally, we validate our model on two freely available captioning benchmarks: Microsoft COCO dataset and Flickr30k dataset. The results demonstrate that our approach achieves state-of-the-art performance and outperforms many of the existing approaches.

Learning the node-to-community partition of network (aka. community detection) lies at the core of network analysis, which have proven to be successful for various tasks. How to improve community detection results via using various features is a challenging task. Existing community detection models integrate links and attributes information of each node in a comprehensive manner, towards a single aspect to represent the node's general role. Yet, a real-world entity could be multifaceted, where it connects to different domains due to different motives that are not necessarily correlated. In this paper, we propose a practical framework for community detection based on attribute and structure information for large networks, namely, Non-exhaustive overlapping community detection in attributed networks via two-stage Maximum likelihood estimation, (for short NotMle). The attribute information can be effectively captured and transformed by attribute embedding to encode the combination with structure information. Then, the link strength among communities is designed to adjust the impact of these structural information on community generation based on the contribution of the structure to the clusters, and the node assignment allow for the nature of the real network (overlapping and outliers). Experimental results on several benchmarks demonstrate encouraging results while learning an interpretable latent structure.

Automatically generating a human-like description for a given image is a potential research in artificial intelligence, which has attracted a great of attention recently. Most of the existing attention methods explore the mapping relationships between words in sentence and regions in image, such unpredictable matching manner sometimes causes inharmonious alignments that may reduce the quality of generated captions. In this paper, we make our efforts to reason about more accurate and meaningful captions. We first propose word attention to improve the correctness of visual attention when generating sequential descriptions word-by-word. The special word attention emphasizes on word importance when focusing on different regions of the input image, and makes full use of the internal annotation knowledge to assist the calculation of visual attention. Then, in order to reveal those incomprehensible intentions that cannot be expressed straightforwardly by machines, we inject external knowledge extracted from knowledge graph into the encoder-decoder framework to facilitate meaningful captioning. We validate our model on two freely available captioning benchmarks: Microsoft COCO dataset and Flickr30k dataset. The results demonstrate that our approach achieves state-of-the-art performance and outperforms many of the existing approaches.

Part 4: Social Computing; International audience; It turns out that the Stochastic Blockmodel (SBM) and its variants can successfully accomplish a variety of tasks, such as discovering community structures. Note that the main limitations are inferencing high time complexity and poor scalability. Our effort is motivated by the goal of harnessing their complementary strengths to develop a scalability SBM for graphs, that also enjoys an efficient inference process and discovery interpretable communities. Unlike traditional SBM that each node is assumed to belong to just one block, we wish to use the node importance to also infer the community membership(s) of each node (as it is one of the goals of SBMs). To this end, we propose a multi-stage maximum likelihood strategy for inferring the latent parameters of adapting the Stochastic Blockmodels to Overlapping Community Detection (OCD-SBM). The intuitive properties to build the model, is more in line with the real-world network to reveal the hidden community structural characteristics. Particularly, this enables inference of not just the node’s membership into communities, but the strength of the membership in each of the communities the node belongs to. Experiments conducted on various datasets verify the effectiveness of our model.

In the relationship extraction task of NLP, how to effective use of the rich structural information on the dependency tree is a challenging research problem. To better learn the dependency relationship between nodes, we address the relationship extraction task by capturing rich contextual dependencies based on the attention mechanism, and using distributional reinforcement learning to generate optimal relation information representation. Unlike using an attention mechanism to effectively make use of relevant information, we propose a Dual Attention Graph Convolutional Network (DAGCN) to adaptively integrate local features with their global dependencies. Specifically, we append two types of attention modules on top of GCN, which model the semantic interdependencies in spatial and relational dimensions respectively. The position attention module selectively aggregates the feature at each position by a weighted sum of the features at all positions of nodes internal features. Similar features would be related to each other regardless of their distances. Meanwhile, the relation attention module selectively emphasizes interdependent node relations by integrating associated features among all nodes. We sum the outputs of the two attention modules and use reinforcement learning to predict the classification of nodes relationship to further improve feature representation which contributes to more precise extraction results. The results on the TACRED and SemEval datasets show that the model can obtain more useful information for relational extraction tasks, and achieve better performances on various evaluation indexes.

Collaborative Filtering is a technique that automatically predicts the interest of an active user by collecting rating information from other similar users or items. However, existing works either leave out the user reliability or cannot capture the implicit correlation between two users who are similar but not socially connected. Besides, they also take the trust value between users is either 0 or 1, thus will degenerate the prediction accuracy. In this work, we propose a social collaborative filtering algorithm via leveraging multisource information in matrix factorization, which aims to solve the above problems via employing both rating records and users’ social network information. To this end, we design a novel recommendation framework, which consists of two fundamental components: computation of user reliability and implicit interaction, designing of user-shared feature space. In particular, the latter performs matrix factorization based on user reliability. Finally, a reasonable objective function is designed to synthetically consider the multisource information of users, such that high-quality recommendation can be produced. Experimental results demonstrate that the proposed model achieves better accuracy than other counterpart recommendation techniques.

Community detection algorithms are the basic tools for discovering the internal structure and organizational principles of a community. Ranging from model-based and optimization-based methods, existing efforts typically consider two sources of information, i.e. network structure and node attributes, to obtain communities with both denser network structure and similar attribute information. We argue that an inherent drawback of such methods is that, different impacts of different sources, is ignored during the clustering process. Besides, some existing community detection algorithms typically consider two sources of information but they cannot automatically determine the relative importance between them to reveal subspaces. As such, the detected communities may be unsatisfactory.

Classical community search methods aim to detect local communities containing a set of sample nodes provided by users, which have been wildly studied in recent years. Existing efforts on community search have mainly detected communities where the sample nodes are located. Nevertheless, they may fail to capture communities without sample nodes but are similar with user’s preference deduced from the given sample nodes. We argue that community search should take user’s preference into account during searching process, steering the algorithm to capture more interesting parts of the entire attributed graph. In this work, we propose a community search model that is capable of finding multiple target communities with few given sample nodes and simultaneously identify outliers in attributed network. The model is termed as Searching Target Communities with Outliers (STCO), which collaborates user’s preference into the process of searching to find interesting clusters of the entire network. Particularly, we specify two STCO methods, named STCOE and STCOT, based on two strategies of exploring sample node candidates, respectively. The average partition similarity is defined on the expanded candidate node set to infer the attribute subspace as user’s latent interest. And then, multiple communities and outliers in the whole network are detected via fractional-core and structural constraints. We demonstrate the effectiveness and efficiency of our model on several synthetic and real-world attributed networks with different scales and subjects.

Most of the existing image captioning models mainly use global attention, which represents the whole image features, local attention, representing the object features, or a combination of them; there are few models to integrate the relationship information between various object regions of the image. But this relationship information is also very instructive for caption generation. For example, if a football appears, there is a high probability that the image also contains people near the football. In this article, the relationship feature is embedded into the global-local attention to constructing a new Pyramid Attention mechanism, which can explore the internal visual and semantic relationship between different object regions. Besides, to alleviate the exposure bias problem and make the training process more efficient, we propose a new method to apply the Generative Adversarial Network into sequence generation. The greedy decoding method is used to generate an efficient baseline reward for self-critical training. Finally, experiments on MSCOCO dataset show that the model can generate more accurate and vivid captions and outperforms many recent advanced models in various prevailing evaluation metrics on both local and online test sets.

Computational modeling of neuronal morphologies is significant for understanding structure-function relationships and brain information processing in computational neuroscience. Using a gene regulatory network model, an evolutionary developmental approach is presented for efficient generation of 3D virtual neurons. This approach describes the developmental process of dendritic morphologies by locally inter-correlating morphological variables which can be represented by the dynamics of gene expression. Then, the multi-objective evolutionary algorithm with gene segmental duplication and divergence operators is applied to evolve the virtual neurons, which aims at generating virtual neurons that are as good as the experimentally traced real neurons in terms of statistical morphological measurements. We experimentally generated motoneurons and statistically compared between the real neurons and the generated virtual neurons by measuring a series of emergent morphological features. The results show that the generated virtual neurons are seemingly realistic, accurate, and further suggest that this approach is an efficient tool for understanding neural development and investigating the relation of neuronal structure to function in particular.

Microblog clustering is very important in many web applications. However, microblogs do not provide sufficient word occurrences. Meanwhile the limited length of these messages prevents traditional text clustering approaches from being employed to their full potential. To address this problem, in this paper, we propose a novel semi-supervised learning scheme fully exploring the semantic information to compensate for the limited message length. The key idea is to explore term correlation data, which well captures the semantic information for term weighting and provides greater context for microblogs. We then formulate microblog clustering problem as a semi-supervised non-negative matrix factorization co-clustering framework, which takes advantage of both prior domain knowledge of data points (microblogs) in the form of pair-wise constraints and category knowledge of features (terms). Our approach not only greatly reduces the labor-intensive labeling process, but also deeply exploits hidden information from microblog itself. Extensive experiments are conducted on two real-world microblog datasets. The results demonstrate the effectiveness of the proposed approach which produces promising performance as compared to state-of-the-art methods.

The intelligent information processing of the standard Zhuang language spoken mainly in Southern China is presently in its infancy, and lacks a well-defined language corpus and automatic part-of-speech tagging methods. Therefore, this study proposes an adversarial part-of-speech tagging method based on reinforcement learning, which solves the problems associated with a lack of a language corpus, time-consuming laborious manual marking, and the low performance of machine marking. Firstly, we construct a markup dictionary based on the grammatical characteristics of standard Zhuang and the Penn Chinese Treebank. Secondly, a dependency syntax analysis is applied for constructing the semantic information feature vectors of sentences, and long short-term memory is adopted as the policy network architecture to enhance available information using recurrent memory, and a conditional random field is employed as the discriminant network to perform label inference with global normalization. Finally, we use reinforcement learning as the model framework, target parts of speech as the feedback of the environment, and then obtain the optimal policy through adversarial learning. The results show that the combination of reinforcement learning and adversarial network alleviates the dependence of the model on the training corpus to some extent, and can quickly and effectively expand the scale of the annotation dictionary for the Zhuang language, thereby obtaining better labeling results.

Short text representation (STR) has attracted increasing interests recently with the rapid growth of Web and social media data existing in short text form. In this paper, we present a new method using an improved semantic feature space mapping to effectively represent short texts. Firstly, semantic clustering of terms is performed based on statistical analysis and word2vec, and the semantic feature space can then be represented via the cluster center. Then, the context information of terms is integrated with the semantic feature space, based on which three improved similarity calculation methods are established. Thereafter the text mapping matrix is constructed for short text representation learning. Experiments on both Chinese and English test collections show that the proposed method can well reflect the semantic information of short texts and represent the short texts reasonably and effectively.

Measuring the similarity between short texts is made difficult by the fact that two texts that are semantically related may not contain any words in common. In this paper, we propose a novel short text similarity measure which aggregates coupled semantic relation (CSR) and strong classification features (SCF) to provide a richer semantic context. On the one hand, CSR considers both intra-relation (i.e. co-occurrence of terms based on the modified weighting strategy) and inter-relation (i.e. dependency of terms via paths that connect linking terms) between a pair of terms. On the other hand, Based on SCF for similarity measure is established based on the idea that the more similar two texts are, the more features of strong classification they share. Finally, we combine the above two techniques to address the semantic sparseness of short text. We carry out extensive experiments on real world short texts. The results demonstrate that our method significantly outperforms baseline methods on several evaluation metrics.

In this paper, we propose a community search scheme via attribute subspace. This method utilizes not only network structure but also node attributes within a certain subspace to quantify a community from the perspective of both internal consistency and external separability, which is able to capture a user preferred community. Firstly, the attributes similarity and neighborhood information of nodes are combined, and the center node set of the target community can be obtained by extending the sample node given by the user with its neighbors. Secondly, an attribute subspace calculation method with entropy weights is established based on the center node set, and the attribute subspace of the community can thus be deduced. Finally, the community quality, which is the combination of internal connectivity and external separability is defined, based on which the target community with user’s preference can be detected. Experimental results on both synthetic network and real-world network datasets demonstrated the efficiency and effectiveness of the proposed algorithm.

In this paper, we present an enhancing graph-based keywords extraction method with node association (GKENA), which strengths graph-based keywords extraction approaches by applying strong association rule mining and unifying three different node attributes into a single framework. Specifically, we regard one single document as a sequential transaction dataset, and apply an efficient algorithm to exploit closed frequent sets and the strong association rules are generated to represent the correlations among multiple terms for association graph construction. Each graph node represents combinations of two or more terms and three node attributes (i.e. graph structure, node semantics and associations) are unified to transfer extra node information into a Markov chain model to obtain the ranking. Besides, in order to avoid the semantic overlapping among top ranked candidate words, a trustworthy clustering algorithm is employed and the center word in each cluster is selected to construct the keywords sets. Our experiments on both Chinese and English datasets indicate that GKENA can boost the quality of keywords in graph-based keywords extraction techniques.

Research on autonomous driving is requiring more affordance information from the environment, which can be efficiently used by the subject vehicle. Road layout parsing is fundamentally important to this, and has attracted deserved attention on the urban road net where traffic signs are well established. However, for the unmarked traffic scenes, like campus and residence community, the large variation on individual surroundings leads to a reduced attention on public research and a lack of both available dataset and generic methods. To explore this issue, we present a novel automatic annotation method to analyze road semantics, which regards the prior trajectories of a vehicle as multi-dimensional space series and extends the traditional time series methods to the space domain to process the data. In order to achieve this, we first propose Incremental Dynamic Space Warping (IDSW) to synchronize space series and extract statistical road feature, then a Bayesian nonparametric method based on the Hierarchical Dirichlet Process-Hidden Markov Model (HDP-HMM) is introduced to generate road semantic sections without explicitly specifying the number of class. Experiment results demonstrate the proposed method can produce road semantics in different level and annotate not only the physical road position but also the vision observations: based on the extracted road modes, the method can annotate corresponding vision image with a positive class label and a rough distance to the road intersection.

© 2017 Elsevier B.V. There has been a growing interest in the research of semantic labeling on scenes represented by 3D point clouds. A fundamental issue that has been largely ignored is the unavoidable presence of unknown objects and the lack of effective ways of dealing with them. Traditional methods usually label unknown objects as one of the pre-trained classes which is either a meaningful target class or a defined unknown class that collectively refers to all uninterested objects. Due to the fact that the class of unknown in essence is a collection of many unseen or uninterested classes, in which the in-class variation is significant and less manageable. It is challenging to solve the unknown problem in a pre-trained manner. In order to advance the research on semantic labeling with the presence of unknown objects, this study investigates the feasibility of adopting an open-set approach, i.e. train a model without unknown objects and reject them accurately in the test. In this paper, we propose a method that exploits the conflict of different labeling results in order to withstand the negative effect of unknown objects. The proposed framework relies on a Conditional Random Field (CRF) to capture inherent spatial relationships and appearance similarities between objects, and employs a Probability of Inclusion Support Vector Machine (PISVM) to estimate an unknown likelihood for each training class. The probabilistic outputs from both CRF and PISVM are then proposed to be combined under the Dempster Shafer theory for conflict measurement and unknown rejection. The novelty lies in that the method encodes both contextual constrains and unknown likelihood for performance enhancement. Comprehensive experimental results on publicly available data sets are presented to show the negative effects of unknown objects and the improvements on labeling accuracy achieved by the proposed method.

In this paper, we present an overlapping microblog community detecting algorithm using a new partition criterion. The new partition criterion considers the explicit and implicit interest of users, together with the user’s attention and practical application. Users containing certain tag in the whole community are extracted as a temporary group and the quality value is calculated under the current partition. The most appropriate core tag is selected and the corresponding group is then updated until certain requirements are satisfied. The community detected by this algorithm share common core tags and the partition results can be revised. Experimental results show that the method is effective and has practical significance.

Using traditional method to extract semantic relations between words hardly applied to micro-blog, which make finding hot event not sensitive. We propose a new method based on restart random walk and Modularity. The semantic relation between items is calculated by conducting the restart random walk iteratively on graph, and then the semantic correlation matrix is constructed. Next, the idea of Modularity is introduced to design algorithm for word clustering, which make a series of micro-blog hot events obtain. The experimental results show that our method has a higher accuracy compared with the kindred method, and hot events could be detected effectively.

© 2016 TCCT. In recent years, there has been a growing interest in the research of semantic labeling of indoor scenes represented by 3D point clouds. A fundamental problem that has largely been oversighted in the current research is the way of dealing with the unknown class which collectively includes all the objects that are of no interest to the application developer. In the training stage, these objects are either completely removed or labeled as unknown, resulting in a trained model which is not fully and fairly exposed to the actual sample space. In the test stage, the unknown objects are naturally present and provided to the classifier, causing a significant drop of the classification accuracy - usually 20%∼30%. Simply improving the features or the classifier will not address the root cause problem. In this paper, we propose a labeling framework combining both Conditional Random Field (CRF) and PI-SVM to specifically solve the problem caused by the unknown class. First, we use a CRF to model the contextual relations in the 3D space, for which the parameters for both node potential and edge potential are learned from training data. Then, we make use of the rejection strategy of the PI-SVM, which estimates an unnormalized probability for each class. Finally, we reinforce the result of CRF with the belief provided by the PI-SVM, and the labeling result is based on the agreement of the two classifiers. The proposed method takes advantage of the global optimization of CRF and the advantage of unknown rejection of PI-SVM. Experimental results on publicly available data set show that this method has improved the classification accuracy by 10.7% given the accuracy drop of 19.23% caused by the unknown.

Traditional hot topic detection algorithms cannot show its optimal performance on microblogs for their inherent flaws in constructing short-text representation model, implementing the core algorithm in large corpus with short time and evaluating the algorithms' qualities during the process of detecting hot topics. In this paper, a novel method for detecting hot topics in microblogs is presented. This approach takes advantage of a probabilistic correlation-based representation measure in order to ensure a dense and low-dimension microblog representation matrix. Besides, we take the clustering as an optimization problem and introduce a discrete particle swarm optimization (DPSO) to simplify the clustering process to detect topics. Furthermore, the clustering quality evaluation criteria is adopted as the optimization objective function for topic detection which can evaluate the algorithms' qualities after each iteration. Experimental results with corpora containing more than 148,000 twitters show that our algorithm is an effective hot topic detection method for microblog.

A short text feature extension method combining statistical information and semantic similarity is proposed,Firstly, After defining the contribution of word, mutual information, an associated word-pairs set is generated by comparing the value of mutual information with threshold, then it is taken as the query words set to search for HowNet. For each word-pairs, senses are found in knowledge base HowNet, and semantic similarity of query word-pairs are calculated. Common sememe satisfied condition is added into the original term vector as extended feature, otherwise, semantic relationship is computed and the corresponding sememe is expanded into feature set. The above process is repeated, an extended feature set is finally obtained. Experimental results show the effectiveness of our method.

Classification of short text is challenging due to its severe sparseness and high dimension, which are typical characteristics of short text. In this paper, we propose a novel approach to classify short texts based on both lexical and semantic features. Firstly, the term dictionary is constructed by selecting lexical features that are most representative words of a certain category, and then the optimal topic distribution from the background knowledge repository is extracted via Latent Dirichlet Allocation. The new feature for short text is thereafter constructed. The experimental results show that our method achieved significant quality enhancement in terms of short text classification.

In order to improve users' experience it is necessary to recommend valuable and interesting content for users. A tag probability correlation based microblog recommendation method TPCMR is presented via analyzing microblog features and the deficiencies of existing microblog recommendation algorithm. Firstly, our method takes advantage of the probability correlation between tags to construct the tag similarity matrix. Then the weight of the tag for each user is enhanced based on the relevance weighting scheme and the user tag matrix can be constructed. The matrix is updated using the tag similarity matrix, which contains both the user interest information and the relationship between tags and tags. Experimental results show that the algorithm is effective for microblog recommendation.

Semi-supervised document clustering, which takes into account limited supervised data to group unlabeled documents into clusters, has received significant interest recently. Because of getting supervised data may be expensive, it is important to get most informative knowledge to improve the clustering performance. This paper presents a semi-supervised document clustering algorithm and a new method for actively selecting informative instance-level constraints to get improved clustering performance. The semi- supervised document clustering algorithm is a Constrained DBSCAN (Cons-DBSCAN) algorithm, which incorporates instance-level constraints to guide the clustering process in DBSCAN. An active learning approach is proposed to select informative document pairs for obtaining user feedbacks. Experimental results show that Cons-DBSCAN with our proposed active learning approach can improve the clustering performance significantly when given a relatively small amount of constraints.

In this paper, we present a microblog recommendation algorithm based on multi-tag correlation. Firstly, a tag retrieval strategy is designed to add tags for unlabeled users, the initial user-tag matrix is then constructed and user-tag weights are set. In order to represent user interests accurately, we fully investigate the associations between the tags. Both inner and outer correlation between tags are defined to conquer the problem of sparsity of user-tag matrix. The user interests can then be decided and microblogs can be recommended to users. Experimental results show that the algorithm is effective for microblog recommendation.

In this paper, we present a semi-supervised clustering method for microblog in which both word-level and microblog document-level constraints are automatically generated totally based on statistical information rather than any kind of external knowledge. The key idea is first to explore term correlation data, which investigates both inter and intra correlation of words, and the initial similarity between words can therefore be deduced. And then an iterative method is established to calculate both word similarity and microblog similarity. The mechanism of incorporating dual constraints is presented based on word similarity and microblog similarity. We then formulate short text clustering problem as a non-negative matrix factorization based on dual constraints. Empirical study of two real-world dataset shows the superior performance of our framework in handling noisy and microblogs.

In order to face the challenges of feature sparsity of short text messages for microblog hot topic detection, in this paper, we first explore the relation between terms, and then build term correlation matrix which is much denser than term-document matrix. Symmetric non-negative matrix factorization (SNMF) on term correlation matrix is applied to obtain term-topic matrix. Finally, we formulated the topic learning problem as probabilistic Latent semantic analysis (pLSA) on term-topic matrix. Besides, this paper also presents the definition of heat and mechanism of sorting the topics. Experiments show that our method can effectively cluster topics and be applied to microblog hot topic detection.

Microblog has the characteristic of short length, complex structure and words deformation. In this paper, a two stage clustering algorithm based on probabilistic latent semantic analysis (pLSA) and K-means clustering (K-means) is proposed. Besides, this paper also presents the definition of popularity and mechanism of sorting the topics. Experiments show that our method can effectively cluster topics and be applied to microblog hot topic detection.

Textual entailment has been proposed as a unifying generic framework for modeling language variability and semantic inference in different Natural Language Processing (NLP) tasks. This paper presents a novel statistical method for recognizing Chinese textual entailment in which lexical, syntactic with semantic matching features are combined together. In order to solve the problems of syntactic tree matching difficulty and tree structure errors caused by Chinese word segmentation, the method firstly clips the syntactic trees into minimum information trees and then computes syntactic matching similarity on them. All features will be used in a voting style under different machine learning methods to predict whether the text sentence can entail the hypothesis sentence in a text-hypothesis pair. The experimental results show that the feature on changing structure of syntactic tree is effective and efficient in Chinese textual entailment.

Clustering microblogs is very important in many web applications. In this paper, we propose a semi-supervised Nonnegative Matrix Factorization clustering method based on term correlation. The key idea is to explore term correlation data, which well captures the semantic information for term weighting. We then formulate microblog clustering problem as a non-negative matrix factorization using word-level constraints. Empirical study of real-world dataset shows the superior performance of our framework in handling noisy and short microblogs.

In this paper, we present a scalable implementation of a topic modeling (Adaptive Link-IPLSA) based method for online event analysis, which summarize the gist of massive amount of changing tweets and enable users to explore the temporal trends in topics. This model also can simultaneously maintain the continuity of the latent semantics to better capture the time line development of events. With the help of this model, users can quickly grasp major topics in these twitters. The preliminary results show that our method leads to more balanced and comprehensive improvement for online event detection compared to benchmark approaches. Additionally our algorithm is computationally feasible in near real-time scenarios making it an attractive alternative for capturing the rapidly changing dynamics of microblogs.

This paper presents a novel time window based hot topic extraction model for news stream. The model considers both pervasiveness and burst feature of topic terms. Pervasiveness is evaluated by terms' occurrences reported from different channels and burst is assessed by terms' abnormal occurrence frequencies from different time intervals. An energy ratio threshold based approach for burst detection is adopted and time window is introduced for news text stream analysis. TF-PDF is then combined to weigh the terms. Experiment results demonstrate that our model is effective in topic extraction for news texts.

Semi-supervised clustering is often viewed as using labeled data to aid the clustering process However, existing algorithms fail to consider dual constraints between data points (e.g documents) and features (e.g words) To address this problem, in this paper, we propose a novel semi-supervised document co-clustering model OSS-NMF via orthogonal nonnegative matrix tri-factorization Our model incorporates prior knowledge both on document and word side to aid the new word-category and document-cluster matrices construction Besides, we prove the correctness and convergence of our model to demonstrate its mathematical rigorous Our experimental evaluations show that the proposed document clustering model presents remarkable performance improvements with certain constraints.

The content and structure of linked information such as sets of web pages or research paper archives are dynamic and keep on changing. Even though different methods are proposed to exploit both the link structure and the content information, no existing approach can effectively deal with this evolution. We propose a novel joint model, called Link-IPLSI, to combine texts and links in a topic modeling framework incrementally. The model takes advantage of a novel link updating technique that can cope with dynamic changes of online document streams in a faster and scalable way. Furthermore, an adaptive asymmetric learning method is adopted to freely control the assignment of weights to terms and citations. Experimental results on two different sources of online information demonstrate the time saving strength of our method and indicate that our model leads to systematic improvements in the quality of classification and link prediction.

This paper presents a framework that actively selects informative documents pairs for semi-supervised document clustering. The semi-supervised document clustering algorithm is a Constrained DBSCAN (Cons-DBSCAN), which incorporates instance-level constraints to guide the clustering process in DBSCAN. By obtaining user feedbacks, our proposed active learning algorithm can get informative instance level constraints to aid clustering process. Experimental results show that Cons-DBSCAN with the proposed active learning approach can provide an appealing clustering performance.

Automatic image annotation has become an important and challenging problem due to the existence of the semantic gap. In this paper, we present an approach based on probabilistic latent semantic analysis (PLSA) to achieve the task. In order to model training data precisely, an image is firstly represented as a bag of visual words, then a probabilistic structure with two PLSA models is employed to capture semantic information from visual and textual modalities respectively. Furthermore, an adaptive asymmetric learning approach is proposed to fuse the aspects of these two models. For each image document, the distribution over aspects of different models is fused by multiplying different weights, which are determined by the entropy of the feature distribution. Consequently, the two models are linked with the same distribution over all aspects. This structure can predict semantic annotation well for an unseen image because it associates visual and textual modalities properly. We compare our approach with several previous approaches on a standard Corel dataset. The experimental results show that our approach performs more effectively and accurately.