Abstract: With the increasing development of electronic technology, traditional paper-driven medical systems have been converting to efficient electronic records that can be easily checked and transmitted. However, due to system updating and equipment failure, missing data problems are very common in the healthcare field. Health data can help people evaluate their health status and adjust their fitness. Therefore, predicting missing health data is a current pressing task. There are two challenges when predicting missing data: (1) people’s health data are complex. The data contain multiple data types (such as continuous data, discrete data and Boolean data) and (2) privacy issues are raised at the edge because huge amounts of health data are published while the edge devices can only provide limited computing and storage resources. Therefore, a novel multitype health data privacy-aware prediction approach based on locality-sensitive hashing is proposed in this paper. Through locality-sensitive hashing, our proposed method can realize a good tradeoff between prediction accuracy and privacy preservation. Finally, through a set of experiments deployed on the WISDM dataset, we verify the validity of our approach in dealing with multitype data and attaining user privacy.

Abstract: The Internet of Things (IoT) is a rapidly emerging technology in the consumer and industrial market. This technology has the potential to radically transform the consumer experience, as it will change our daily scenes, starting from the way we drink coffee to how smart objects interact with industrial applications. Such rapid development and deployment face multifarious challenges, including the sheer amount of data generated, network scale, network heterogeneity, as well as security and privacy concerns. In recent years, Distributed Denial-of-Service (DDoS) attacks in IoT networks are considered one of the growing challenges that need to be shed light on. DDoS attacks utilize the limited resources in IoT devices, such as storage limitation and network capacity, that cause this issue in the IoT application. This paper comprehensively reviews the attacks that can lead to DDoS, which eventually will cause serious damage to existing systems. Additionally, the paper investigates the available solutions used to counter these attacks and explore their limitations from the perspective of the constrained device. Furthermore, a detailed analysis of the existing solution placement was implemented, including heterogeneity and their performance for IoT based networks. Finally, the paper will reveal and discuss interesting research direction on the future IoT security and current trends.

Abstract: Exploitation time is an essential factor for vulnerability assessment in cybersecurity management. In this work, we propose an integrated consecutive batch learning framework to predict the probable exploitation time of vulnerabilities. To achieve a better performance, we combine features extracted from both vulnerability descriptions and the Common Vulnerability Scoring System in the proposed framework. In particular, we design an Adaptive Sliding Window Weighted Learning (ASWWL) algorithm to tackle the dynamic multiclass imbalance problem existing in many industrial applications including exploitation time prediction. A series of experiments are carried out on a real-world dataset, containing 24,413 exploited vulnerabilities disclosed between 1990 and 2020. Experimental results demonstrate the proposed ASWWL algorithm can significantly enhance the performance of the minority classes without compromising the performance of the majority class. Besides, the proposed framework achieves the most robust and state-of-the-art performance compared with the other five consecutive batch learning algorithms.

Abstract: As a constructive mode of information sharing, collaboration and communication, social media platforms offer users with limitless opportunities. The same hypermedia can be transposed into a synthetic and toxic milieu that provides an anonymous, destructive pedestal for online bullying and harassment. Automatic cyberbullying detection on social media using synthetic or real-world datasets is one of a proverbial natural language processing problem. Analyzing a given text requires capturing the existent semantics, syntactic and spatial relationships. Learning representative features automatically using deep learning models efficiently captures the contextual semantics and word order arrangement to build robust and superlative predictive models. This work puts forward a hybrid model, Bi-GRU-Attention-CapsNet (Bi-GAC), that benefits by learning sequential semantic representations and spatial location information using a Bi-GRU with self-attention followed by CapsNet for cyberbullying detection in the textual content of social media. The proposed Bi-GAC model is evaluated for performance using F1-score and ROC-AUC curve as metrics. The results show a superior performance to the existing techniques on the benchmark Formspring.me and MySpace datasets. In comparison to the conventional models, an improvement of nearly 9% and 3% in F-score is observed for MySpace and Formspring.me dataset respectively.

Abstract: The ubiquitous Internet of Things (IoTs) devices spawn growing mobile services of applications with computationally-intensive and latency-sensitive features, which increases the data traffic sharply. Driven by container technology, microservice is emerged with flexibility and scalability by decomposing one service into several independent lightweight parts. To improve the quality of service (QoS) and alleviate the burden of the core network, caching microservices at the edge of networks empowered by the mobile edge computing (MEC) paradigm is envisioned as a promising approach. However, considering the stochastic retrieval requests of IoT devices and time-varying network topology, it brings challenges for IoT devices to decide the caching node selection and microservice replacement independently without complete information of dynamic environments. In light of this, a MEC-enabled di stributed cooperative m icroservice ca ching scheme, named DIMA, is proposed in this paper. Specifically, the microservice caching problem is modeled as a Markov decision process (MDP) to optimize the fetching delay and hit ratio. Moreover, a distributed double dueling deep Q-network (D3QN) based algorithm is proposed, by integrating double DQN and dueling DQN, to solve the formulated MDP, where each IoT device performs actions independently in a decentralized mode. Finally, extensive experimental results are demonstrated that the DIMA is well-performed and more effective than existing baseline schemes.

Abstract: In light of the ability to enable efficient storage and fast query for big data, hashing techniques for cross-modal search have aroused extensive attention. Despite the great success achieved, unsupervised cross-modal hashing still suffers from lacking reliable similarity supervision and struggles with handling the heterogeneity issue between different modalities. To cope with these, in this paper, we devise a new deep hashing model, termed as High-order Nonlocal Hashing (HNH) to facilitate cross-modal retrieval with the following advantages. First, different from existing methods that mainly leverage low-level local-view similarity as the guidance for hashing learning, we propose a high-order affinity measure that considers the multi-modal neighbourhood structures from a nonlocal perspective, thereby comprehensively capturing the similarity relationships between data items. Second, a common representation is introduced to correlate different modalities. By enforcing the modal-specific descriptors and the common representation to be aligned with each other, the proposed HNH significantly bridges the modality gap and maintains the intra-consistency. Third, an effective affinity preserving objective function is delicately designed to generate high-quality binary codes. Extensive experiments evidence the superiority of the proposed HNH in unsupervised cross-modal retrieval tasks over the state-of-the-art baselines.

Abstract: With the rapid development of Information Technology, there exist massive amounts of data available on the Internet, which result in a severe information overload problem. Especially, it becomes more and more challenging but necessary to help users find the contents or services that they really need. To address the problem mentioned above, recommender systems have been developed to exploit user’s historical behavior data and provide personalized services for promoting customer experiences in many fields, such as Point of Interest (POI) applications, multimedia services, and e-commerce websites. Specifically, in POI recommendation, user’s next check-in behaviors depend on both long- and short-term preferences. However, traditional recommendation methods often ignore the dynamic changes of user’s short-term preferences over time, which limits their performance. Besides, many existing methods cannot fully exploit the complex correlations and transitions between POI in check-ins sequences. In this paper, we propose an A ttentive S equential model based on G raph N eural N etwork (ASGNN) for accurate next POI recommendation. Specifically, ASGNN firstly models user’s check-in sequences as graphs and then use Graph Neural Networks (GNN) to learn the informative low-dimension latent feature vectors (embeddings) of POIs. Secondly, a personalized hierarchical attention network is adopted to exploit complex correlations between users and POIs in check-in sequences and capture user’s long- and short-term preferences. Finally, we perform the next POI recommendation via leveraging user’s long- and short-term preferences obtained from their behavior sequences with ASGNN. Extensive experiments are conducted on three real-world check-in datasets, and the results demonstrate that the proposed model ASGNN outperforms baselines, including some state-of-the-art methods.

Abstract: A growing number of web APIs published on the Internet allows mashup developers to discover appropriate web APIs for polishing mashups. Developers often have to manually pick and choose several web APIs from extremely massive candidates, which is a laborious and cumbersome task. Fortunately, recommender system comes into existence. Some approaches perform recommendations in cloud platforms by utilizing historical records of Mashup-API interactions stored in edge nodes. However, many of these methods often pay more attention to recommendation accuracy while ignoring recommendation diversity, i.e., there are usually popular web APIs in recommendation list while most of the other novel web APIs are absent. The poor recommendation diversity may limit the usefulness of the recommendation results due to the lack of novelty. In order to implement an accurate and diversified web API recommendation, a novel MF-based recommendation approach named Div_PreAPI is put forward in this paper. Div_PreAPI integrates a weighting mechanism and neighborhood information into matrix factorization (MF) to implement diversified and personalized APIs recommendations. Finally, we conduct a series of experiments on a real-world dataset. Experimental results show the effectiveness of our proposal.

Abstract: Web API is an efficient and cost-effective method for service-oriented software development, and Mashup is a popular technology which combines multiple services to create more powerful services to address the increasing complexity of business requirements and speed up the software development process. Here, accurate and efficient API recommendation is vital for successful Mashup development. Currently, many existing methods combine various technologies and adopt diverse features, which results in complex models at the cost of higher computational overhead but with very limited improvement on recommendation accuracy. To address such an issue, in this paper, we propose an unsupervised API recommendation method based on deep random walks on knowledge graph. Specifically, we first construct a refined knowledge graph utilizing Mashup-API co-invocation patterns and service category attributes, and then we learn implicit low-dimensional embedding representations of entities from truncated random walks by treating walks as the equivalent of sentences. Meanwhile, to improve the recommendation accuracy, we design an entity bias procedure to reflect different entity preference (namely API-based neighborhood or Mashup-based neighborhood). Finally, we estimate the relevance between Mashup requirements and the existing services (Mashups and APIs) to obtain the API recommendation list. Since the API recommendation results can be obtained through unsupervised feature learning, automatic API recommendation can be provided for Mashup developers in real time. Comprehensive experimental results on a real-world dataset demonstrate that our proposed method can outperform several state-of-the-art methods in both recommendation accuracy and efficiency.

Abstract: Although data analysis and mining technologies can efficiently provide intelligent and personalized services to us, data owners may not always be willing to share their true data because of privacy concerns. Recently, differential privacy (DP) technology has achieved a good trade-off between data utility and privacy guarantee by publishing noisy outputs. Nonetheless, DP still has a risk of privacy leakage when handling correlated data directly. Current schemes attempt to extend DP to publish correlated data, but are faced with the challenge of violating DP or low-level data utility. In this paper, we try to explore the essential cause of this inapplicability. Specifically, we suppose that this inapplicability is caused by the different correlations between noise and original data. To verify our supposition, we propose the notion of Correlation-Distinguishability Attack (CDA) to separate IID (Independent and Identically Distributed) noise from correlated data. Furthermore, taking time series as an example, we design an optimum filter to realize CDA in practical applications. Experimental results support our supposition and show that, the privacy degree of current approaches has a degradation under CDA.

Abstract: Recently, the development of data mining and natural language processing techniques enable the relationship probe between social media and stock market volatility. The integration of natural language processing, deep learning and the financial field is irresistible. This paper proposes a hybrid approach for stock market prediction based on tweets embedding and historical prices. Different from the traditional text embedding methods, our approach takes the internal semantic features and external structural characteristics of Twitter data into account, such that the generated tweet vectors can contain more effective information. Specifically, we develop a Tweet Node algorithm for describing potential connection in Twitter data through constructing the tweet node network. Further, our model supplements emotional attributes to the Twitter representations, which are input into a deep learning model based on attention mechanism together with historical stock price. In addition, we designed a visual interactive stock prediction tool to display the result of the prediction.

Abstract: Aspect-level sentiment classification aims to identify sentiment polarity over each aspect of a sentence. In the past, such analysis tasks mainly relied on text data. Nowadays, due to the popularization of smart devices and Internet services, people are generating more abundant data, including text, image, video, et al. Multimodal data from the same post (e.g., a tweet) usually has certain correlation. For example, image data might has an auxiliary effect on the text data, and reasonable processing of such multimodal data can help obtain much richer information for sentiment analysis. To this end, we propose an aspect-level sentiment classification model by exploring multimodal data with fusion discriminant attentional network. Specifically, we first leverage two memory networks for mining the intra-modality information of text and image, and then design a discriminant matrix to supervise the fusion of inter-modality information. Experimental results demonstrate the effectiveness of the proposed model.

Abstract: Recently, there have been some breakthroughs in graph analysis by applying the graph neural networks (GNNs) following a neighborhood aggregation scheme, which demonstrate outstanding performance in many tasks. However, we observe that the parameters of the network and the embedding of nodes are represented in real-valued matrices in existing GNN-based graph embedding approaches which may limit the efficiency and scalability of these models. It is well-known that binary vector is usually much more space and time efficient than the real-valued vector. This motivates us to develop a binarized graph neural network to learn the binary representations of the nodes with binary network parameters following the GNN-based paradigm. Our proposed method can be seamlessly integrated into the existing GNN-based embedding approaches to binarize the model parameters and learn the compact embedding. Extensive experiments indicate that the proposed binarized graph neural network, namely BGN, is orders of magnitude more efficient in terms of both time and space while matching the state-of-the-art performance.

Abstract: Recommender systems, which are used to predict user requirements precisely, play a vital role in the modern internet industry. As an effective tool with rich semantics, knowledge graphs have recently attracted growing research attention in enhancing recommendation results. By mining multihop relations (i.e., paths) between user-item interactions within a knowledge graph, implicit user preferences and other side information can be clearly revealed. Nevertheless, existing knowledge graph-based recommendation methods have two fundamental limitations. First, the indiscriminate utilization of user-item path sets conveys unclear information and negatively influences explainability. Moreover, obtaining reliable recommendation results with these methods requires large amounts of prior knowledge, which indicates that they show poor performance in terms of accuracy and handling cold-start issues. To address these issues, we propose a novel model called the Path-enhanced Recurrent Network (PeRN). Specifically, PeRN integrates a recurrent neural network encoder with a metapath-based entropy encoder to increase explainability and accuracy and reduce cold-start costs. The recurrent network encoder has a strong ability to represent sequential path semantics in a knowledge graph, while the entropy encoder, as an efficient statistical analysis tool, leverages metapath information to differentiate paths in a single user-item interaction. A path extraction algorithm with a bidirectional scheme is also proposed to make PeRN more feasible. The experimental results on two real-world datasets demonstrate our significant improvements with reasonable explanations, promising accuracy and a minimal amount of prior knowledge compared with several state-of-the-art baselines.

Abstract: If blockchain networks are to become the building blocks of the infrastructure for the future digital economy, then several challenges related to the resiliency and survivability of blockchain networks need to be addressed. The survivability of a blockchain network is influenced by the diversity of its nodes. Trustworthy device-level attestations permits nodes in a blockchain network to provide truthful evidence regarding their current configuration, operational state, keying material and other system attributes. In the current work we review the recent developments towards a standard attestation architecture and evidence conveyance protocols. We explore the applicability and benefits of a standard attestation architecture to blockchain networks. Finally, we discuss a number of open challenges related to node attestations that has arisen due to changing model of blockchain network deployments, such as the use of virtualization and containerization technologies for nodes in cloud infrastructures.

Abstract: Human-AI collaborative decision-making tools are being increasingly applied in critical domains such as healthcare. However, these tools are often seen as closed and intransparent for human decision-makers. An essential requirement for their success is the ability to provide explanations about themselves that are understandable and meaningful to the users. While explanations generally have positive connotations, studies showed that the assumption behind users interacting and engaging with these explanations could introduce trust calibration errors such as facilitating irrational or less thoughtful agreement or disagreement with the AI recommendation. In this paper, we explore how to help trust calibration through explanation interaction design. Our research method included two main phases. We first conducted a think-aloud study with 16 participants aiming to reveal main trust calibration errors concerning explainability in AI-Human collaborative decision-making tools. Then, we conducted two co-design sessions with eight participants to identify design principles and techniques for explanations that help trust calibration. As a conclusion of our research, we provide five design principles: Design for engagement, challenging habitual actions, attention guidance, friction and support training and learning. Our findings are meant to pave the way towards a more integrated framework for designing explanations with trust calibration as a primary goal.

Abstract: Influence maximization, as one of the major problems in Location-based Social Networks (LBSN), aims to determine a subset of influential users to maximize the influence spread through the “word-of-mouth” effect. Although many recent studies have focused on the influence maximization problem in LBSN, a majority part of the concern is shed on the influence spread in the whole network, with an underestimation in the importance of the community structure. In this paper, we propose a Community-based Influence Maximization model to study the influence maximization problem in LBSN, with consideration of both community structure and users’ spatio-temporal behavior. Two community-based algorithms are developed to maximize the influence spread, which encompass two components: 1) detecting communities in LBSN based on users’ mobility; and 2) selecting the most influential individuals based on communities. In the first phase, we calculate the similarity between users according to their historical check-in data and design a Weighted Distance algorithm to detect communities based on the similarity. In the second phase, we select candidates based on local network structure and propose two different methods to calculate the precise influence spread of each candidate based on communities. The extensive experiments over real datasets demonstrate the efficiency and effectiveness of the proposed algorithms.

Abstract: Bipartite graphs, which consist of two different types of entities, are widely used to model many real-world applications. In bipartite networks, (α,β)-core is an essential model to measure the entity engagement. In this paper, we propose and investigate the problem of (α,β)-core minimization, which aims to identify a set of b edges whose deletion can minimize the size of resulting collapsed (α,β)-core. To our best knowledge, this is the first work to investigate the (α,β)-core minimization problem in bipartite graph. We prove the problem is NP-hard and our object function is monotonic but not submodular. Then, we propose a baseline algorithm by invoking the greedy framework. To reduce the computation cost and candidate space, novel pruning techniques are devised. We further develop a group based algorithm to optimize the search. Finally, we conduct comprehensive experiments over 6 real-life bipartite networks to demonstrate the advantages of the proposed techniques.

Abstract: Blockchain technology enables several untrustworthy parties to execute inter-organizational business processes in a tamper-proof manner. Existing approaches are based on smart contract code generation. Although the smart contract immutability guarantees that the process model is followed by each participant, any change in the process model makes the existing process instance smart contract not compliant with the process model and therefore, the deployment of a new smart contract becomes a requirement. In this paper, we propose CoBuP, a decentralized Collaborative Business Process execution architecture using blockchain. To address the inflexibility issue, this paper presents an interpreter of BPMN process models, supporting the instantiating, execution, monitoring of process instances. For this reason, one generic smart contract is deployed once, which helps generate a process instance, create BPMN elements by invoking the process instance predefined functions, and dynamically update the adaptation of a process at run time. For validation purposes, two use cases have permitted to carry out different experiments that demonstrate the technical feasibility of our approach. The obtained results of the experimental evaluation shows that our approach cost is similar or lower than that of the existing approaches.

Abstract: With the rapid development of artificial intelligence and the increasing demand for legal intelligence, using AI methods to predict legal judgments has become a hot spot in recent years. Charge prediction is one of the core tasks of Legal Judgment Prediction (LJP). It aims to predict charge from complicated legal facts, so as to help the court make judgments or provide legal professional guidance to non-professionals. In the field of legalAI, interpretability is crucial compared to others. Reasonable interpretability can eliminate hidden dangers such as gender discrimination and provide support for judges’ decisions. However, how to add the legal theory framework to the modeling to improve the interpretability is a challenge, which has few researches at present. To address this problem, we use Double-layer Criminal System as a guide to build Charge Prediction modeling called DCSCP which aims to predict charges in the criminal law of China. In general, our characteristic is to achieve multi-granularity inference of legal charges by obtaining the subjective and objective elements from the fact descriptions of legal cases. Specifically, our approach is performed in two steps: (1) extract the objective elements from the fact description and use them to generate candidate charges to achieve coarse-grained prediction; (2) extract the subjective elements from the fact description, and design the first-order predicate logic inference to realize the fine-grained charge inference in combination with the candidate charges. Experimental results show that our DCSCP can provide interpretable predictions, and it can maintain performance compared to other state-of-the-art charge prediction models.

Abstract: Patent classification has long been treated as a crucial task to support related services. Though large efforts have been made on the automatic patent classification task, those prior arts mainly focus on mining textual information such as titles and abstracts. Meanwhile, few of them pay attention to the meta data, e.g., the inventors and the assignee company, and the potential correlation via the metadata-based graph has been largely ignored. To that end, in this paper, we develop a new paradigm for patent classification task in the perspective of multi-view patent graph analysis and then propose a novel framework called Patent2vec to learn low-dimensional representations of patents for patent classification. Specifically, we first employ the graph representation learning on individual graphs, so that view-specific representations will be learned by capturing the network structure and side information. Then, we propose a view enhancement module to enrich single view representations by exploiting cross-view correlation knowledge. Afterward, we deploy an attention-based multi-view fusion method to get refined representations for each patent and further design a view alignment module to constraint final fused representation in a relational embedding space which can preserve latent relational information. Empirical results demonstrate that our model not only improves the classification accuracy but also improves the interpretability of classifying patents reflected in the multi-source data.

Abstract: Temporal graphs are equipped with entities and the relationships between entities associated with time stamps. Cohesive subgraph mining (CSM) is a fundamental task in temporal graph analysis, which has gathered great research interests. It benefits from reflecting the dynamism of graphs and has many real-world applications. Yet, most existing work focus on the cohesive subgraph detection (CSD) problem, which identifies all the defined subgraphs in the entire temporal graphs. When graph size becoming too large, it is impractical. In this paper, we are the first to concern about the cohesive subgraph search (CSS) problem in large temporal graphs. In specific, given a query vertex, we are seeking the continual densely connected subgraph including the query vertex. To this end, (1) we model the cohesive subgraph in temporal graphs as a (𝜃,τ)-continual k-core and prove its NP-hardness; (2) we develop two exact algorithms based on different vertex enumeration strategies, called Exact-VD and Exact-VE, respectively. Exact-VD uses depth-first search to find the target subgraphs in a top-down way by gradually deleting vertices from the current subgraph; while Exact-VE starts from the query vertex and continuously expands the ranked vertices in the candidate group until reaching the target subgraphs. Meanwhile, several elegant pruning rules are designed to reduce the search space; (3) to further speed up, we propose an efficient approximate local search method, called Approx-LS, which greedily expands the current subgraph guided by the developed heuristic functions until identifying the results. Comprehensive experiments on four real-life datasets verify the efficiency and effectiveness of our proposed approaches.

Abstract: Heterogeneous information network (HIN) embedding represents heterogeneous nodes as vectors in the low-dimensional space. Meta-path is used to measure the nodes similarity to guide HIN embedding. Existing works assume that different meta-paths share the same semantic space and directly fuse the different mate-paths for node similarity calculation. This ignores the incompatibility of different meta-paths, which cannot reflect the real relationship between nodes. To solve the problems of existing works, a novel S emantic-A ware H IN E mbedding (SAHE) is proposed to fuse incompatible meta-paths for node similarity measurement. The key idea of the proposed method is to measure the relative similarity relationship on each meta-path in its own semantic space, and aggregate these similarity relationships to obtain the node similarity to calculate HIN embedding. The kendall tau distance is used to aggregate the different similarity relationship in multiple semantic spaces. The semantic preference is extracted as a constraint to optimize the aggregated similarity matrix. The Kullback-Leibler Divergence (KL Divergence) is used to learn nodes embedding by measuring the node similarity distribution in the embedded space. Experiments on three real HIN datasets verify that the superiority of the proposed model is superior to other state-of-the-art methods on the node classification and the node clustering tasks.

Abstract: Next Point-of-interest (POI) recommendation has been recognized as an important technique in location-based services, and existing methods aim to utilize sequential models to return meaningful recommendation results. But these models fail to fully consider the phenomenon of user interest drift, i.e. a user tends to have different preferences when she is in out-of-town areas, resulting in sub-optimal results accordingly. To achieve more accurate next POI recommendation for out-of-town users, an adaptive attentional deep neural model HOPE is proposed in this paper for modeling user’s out-of-town dynamic preferences precisely. Aside from hometown preferences of a user, it captures the long and short-term preferences of the user in out-of-town areas using “Asymmetric-SVD” and “TC-SeqRec” respectively. In addition, toward the data sparsity problem of out-of-town preference modeling, a region-based pattern discovery method is further adopted to capture all visitor’s crowd preferences of this area, enabling out-of-town preferences of cold start users to be captured reasonably. In addition, we adaptively fuse all above factors according to the contextual information by adaptive attention, which incorporates temporal gating to balance the importance of the long-term and short-term preferences in a reasonable and explainable way. At last, we evaluate the HOPE with baseline sequential models for POI recommendation on two real datasets, and the results demonstrate that our proposed solution outperforms the state-of-art models significantly.

Abstract: Attributed network embedding has attracted plenty of interest in recent years. It aims to learn task-independent, low-dimensional, and continuous vectors for nodes preserving both topology and attribute information. Most of the existing methods, such as random-walk based methods and GCNs, mainly focus on the local information, i.e., the attributes of the neighbours. Thus, they have been well studied for assortative networks (i.e., networks with communities) but ignored disassortative networks (i.e., networks with multipartite, hubs, and hybrid structures), which are common in the real world. To model both assortative and disassortative networks, we propose a block-based generative model for attributed network embedding from a probability perspective. Specifically, the nodes are assigned to several blocks wherein the nodes in the same block share the similar linkage patterns. These patterns can define assortative networks containing communities or disassortative networks with the multipartite, hub, or any hybrid structures. To preserve the attribute information, we assume that each node has a hidden embedding related to its assigned block. We use a neural network to characterize the nonlinearity between node embeddings and node attributes. We perform extensive experiments on real-world and synthetic attributed networks. The results show that our proposed method consistently outperforms state-of-the-art embedding methods for both clustering and classification tasks, especially on disassortative networks.

Abstract: With growing concern of data privacy violations, privacy preservation processes become more intense. The k-anonymity method, a widely applied technique, transforms the data such that the publishing datasets must have at least k tuples to have the same link-able attribute, quasi-identifiers, values. From the observations, we found that, in a certain domain, all quasi-identifiers of the datasets, can have the same data type. This type of attribute is considered as an Identical Generalization Hierarchy (IGH) data. An IGH data has a particular set of characteristics that could utilize for enhancing the efficiency of heuristic privacy preservation algorithms. In this paper, we propose a data privacy preservation heuristic algorithm on IGH data. The algorithm is developed from the observations on the anonymous property of the problem structure that can eliminate the privacy constraints consideration. The experiment results are presented that the proposed algorithm could effectively preserve data privacy and also reduce the number of visited nodes for ensuring the privacy protection, which is the most time-consuming process, compared to the most efficient existing algorithm by at most 21%.

Abstract: Aspect-based sentiment analysis has received considerable attention in recent years because it can provide more detailed and specific user opinion information. Most existing methods based on recurrent neural networks usually suffer from two drawbacks: information loss for long sequences and a high time consumption. To address such issues, a hybrid attention model is proposed for aspect-based sentiment analysis in this paper, which utilizes only attention mechanisms rather than recurrent or convolutional structures. In this model, a self-attention mechanism and an aspect-attention mechanism are designed for generating the semantic representation at the word and sentence levels respectively. Two auxiliary features of word location and part-of-speech are also explored for the proposed models to enhance the semantic representation of sentences. A series of experiments are conducted on three benchmark datasets for aspect-based sentiment analysis. Experimental results demonstrate the advantage of the proposed models for both efficiency and execution effectiveness.

Abstract: Modelling mix-and-match relationships among fashion items has become increasingly demanding yet challenging for modern E-commerce recommender systems. When performing clothes matching, most existing approaches leverage the latent visual features extracted from fashion item images for compatibility modelling, which lacks explainability of generated matching results and can hardly convince users of the recommendations. Though recent methods start to incorporate pre-defined attribute information (e.g., colour, style, length, etc.) for learning item representations and improving the model interpretability, their utilisation of attribute information is still mainly reserved for enhancing the learned item representations and generating explanations via post-processing. As a result, this creates a severe bottleneck when we are trying to advance the recommendation accuracy and generating fine-grained explanations since the explicit attributes have only loose connections to the actual recommendation process. This work aims to tackle the explainability challenge in fashion recommendation tasks by proposing a novel Attribute-aware Fashion Recommender (AFRec). Specifically, AFRec recommender assesses the outfit compatibility by explicitly leveraging the extracted attribute-level representations from each item’s visual feature. The attributes serve as the bridge between two fashion items, where we quantify the affinity of a pair of items through the learned compatibility between their attributes. Extensive experiments have demonstrated that, by making full use of the explicit attributes in the recommendation process, AFRec is able to achieve state-of-the-art recommendation accuracy and generate intuitive explanations at the same time.

Abstract: Knowledge, in practice, is time-variant and many relations are only valid for a certain period of time. This phenomenon highlights the importance of harvesting temporal-aware knowledge, i.e., the relational facts coupled with their valid temporal interval. Inspired by pattern-based information extraction systems, we resort to temporal patterns to extract time-aware knowledge from free text. However, pattern design is extremely laborious and time consuming even for a single relation, and free text is usually ambiguous which makes temporal instance extraction extremely difficult. Therefore, in this work, we study the problem of temporal knowledge extraction with two steps: (1) temporal pattern extraction by automatically analysing a large-scale text corpus with a small number of seed temporal facts, (2) temporal instance extraction by applying the identified temporal patterns. For pattern extraction, we introduce various techniques, including corpus annotation, pattern generation, scoring and clustering, to improve both accuracy and coverage of the extracted patterns. For instance extraction, we propose a double-check strategy to improve the accuracy and a set of node-extension rules to improve the coverage. We conduct extensive experiments on real world datasets and compared with state-of-the-art systems. Experimental results verify the effectiveness of our proposed methods for temporal knowledge harvesting.