Search

In the past decades, recommendation systems have provided lots of valuable personalized suggestions for the users to address the problem of information over-loaded. Collaborative Filtering (CF) is one of the most commonly applied and successful recommendation approaches, which refers to using the preferences of groups with similar interests to recommend information to other users. Recently, in addition to the traditional matrix factorization techniques, deep learning methods have been proposed to learn more abstract and higher-level representations for recommendation. However, most previous deep recommendation methods learn the higher-level feature representations of users and items through an identical model structure, which ignores the different characteristics of the user-based and item-based data. In addition, the rating matrix is usually sparse which may result in a significant degradation of recommendation performance. To address these problems, we propose a representation learning method with Collaborative Autoencoder for Personalized Recommendation (CAPR for short). In this method, user-based and item-based feature representations are learned by two different autoencoders for capturing different features of the data. Meanwhile, items' attributions are combined into the feature representations with semi-autoencoder for alleviating the sparsity problem. Extensive experimental results confirm the effectiveness of our proposed method compared to other state-of-the-art matrix factorization methods and deep recommendation methods. • Two different autoencoders are used to capture characteristics for users and items. • Manifold regularization is integrated into autoencoder for user's features learning. • The comprehensive experiments evaluate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Abstract: Influence maximization in a social network is to target a given number of nodes in the network such that the expected number of activated nodes from these nodes is maximized. A social network usually exhibits some degree of modularity. Previous research efforts that made use of this topological property are restricted to random networks with two communities. In this paper, we firstly transform the influence maximization problem in a modular social network to an optimal resource allocation problem. We assume that the communities of the social network are disconnected. We then propose a recursive relation for finding such an optimal allocation. We prove that finding an optimal allocation in a modular social network is NP-hard and propose a new dynamic programming algorithm to solve the problem. We name our new algorithm OASNET (Optimal Allocation in a Social NETwork). We compare OASNET with the high degree heuristics, the single degree discount heuristics, and the degree discount heuristics on three real world datasets. Experimental results show that OASNET outperforms comparison heuristics significantly on the independent cascade model when the diffusion probability is greater than a certain threshold. [Copyright &y& Elsevier]

This paper proposes an approach KeyRank to extract proper keyphrases from a document in English. It first searches all keyphrase candidates from the document, and then ranks them for selecting top- N ones as final keyphrases. Existing studies show that extracting a complete keyphrase candidate set that includes semantic relations in context, and evaluating the effectiveness of each candidate are crucial to extract high quality keyphrases from documents. Based on that words do not repeatedly appear in an effective keyphrase in English, a novel keyphrase candidate search algorithm using sequential pattern mining with gap constraints (called KCSP) is proposed to extract keyphrase candidates for KeyRank. And then an effectiveness evaluation measure pattern frequency with entropy (called PF-H) is proposed for KeyRank to rank these keyphrase candidates. Our experimental results show that KeyRank has better performance. Its first component KCSP is much more efficient than a closely related approach SPMW, and its second component PF-H is an effective evaluation mechanism for ranking keyphrase candidates. 1 1 Our two-page extended abstract is published in AAAI 2017 (Wang, Sheng, & Wu, 2017). [ABSTRACT FROM AUTHOR]

Over the past few decades, short text classification has emerged as a critical downstream task in natural language processing (NLP). One crucial classification research issue is how to advance semantic understanding considering the short length, feature sparsity, and high ambiguity in short texts. Recently, prompt-tuning has been proposed to insert a template into the input and convert text classification tasks into equivalent cloze-style tasks. However, among most of the previous approaches, either the crafted template methods are time-consuming and labor-intensive, or automatic prompt generation methods cannot achieve satisfied performance. In this paper, we introduce a novel approach called Soft Knowledgeable Prompt-tuning for short text classification. Our method considers both the template generation and classification performance to construct prompts for label prediction. We employ five different strategies to expand the label words space for modifying soft prompts, and the integration of these strategies is used as the final verbalizer. Despite being automatic, experimental results show that our method achieved more desirable performance even than the crafted template methods, outperforming the state-of-the-art by more than 14 Accuracy points on four well-known benchmarks. • Both the automatic prompt generation and text classification performance are achieved. • 5 different strategies are used to construct verbalizer effectively and efficiently. • The comprehensive experiments evaluate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

Knowledge graph completion (KGC) aims to predict the missing element in a triple based on known triples or facts. Recently, plenty of representation learning methods for KGC have achieved the promising performance, especially ones based on graph neural networks and their variants. Those methods exploit local neighborhood information to update the embedding of target entities. However, the existing works have the following two problems. First, those approaches focus on the representation learning of entities, while the relation representation usually adopts a simple linear transformation, which cannot capture the distinctive semantic intensions of the same relation in different triples. Second, different types of entity information are simply combined together, resulting in the loss of global properties including the type and the global importance of entities, which is prone to cause over-smoothing phenomenon. To address these two problems, we propose a Graph Attention Network with Dynamic Representation of Relations and global information (DRR-GAT) for knowledge graph completion. Specifically, the task of dynamic representation of relations is to learn the distinctive representation of the same relation in different triples. This goal is achieved via a path Transformer. To this end, path Transformer is designed to take the path information as its input, where only those paths from the target entity to the neighborhood relations with the same type as the target relation are considered. Sequentially, the mechanism of global embeddings is incorporated into graph attention network to capture the global information of entities and relations. Experimental performance outperforms the state-of-the-art methods, indicating the effectiveness of our proposed approach. • Graph Attention Network with Dynamic Representation of Relations (DRR-GAT) is proposed. • DRR employs Path-Transformer model for learning distinctive local representation • Global embedding is introduced to capture global features of entities and relations • A new message passing function is designed to capture the local information of entities. [ABSTRACT FROM AUTHOR]

Recent advances in interactive recommender systems (IRS) have received wide attention due to its flexible recommendation strategy and optimization for users' long-term utility. Considering this interaction paradigm of IRS, researchers have made some attempts to incorporate reinforcement learning (RL) models into IRS, because of the excellent ability of RL in long-term optimizing and decision making. However, data sparsity is an intractable problem most IRS urgently need to address. Although a small amount of work has exploited reviews to address data sparsity, they ignored the varying importance of items for modeling the user. In addition, most existing RL-based approaches suffer from decision-making difficulties when the action space becomes large. To solve above problems, in this work, we present a Review-enhanced Deep Reinforcement Learning model (REDRL) for interactive recommendation. Specifically, we utilize text reviews, combined with a pretrained review representation model to acquire item review-enhanced embedding representations. Then we formalize the recommendation problem as a Markov Decision Process (MDP), and exploit deep reinforcement learning (DRL) to model the interactive recommendation. Notably, we introduce a multi-head self-attention technique to capture distinct importance of various items in the sequence behavior, which is overlooked by existing work when modeling the user preference. In this way, we can model long-term dynamic preferences of users accurately and discriminately for comprehensive interactive recommendation. Moreover, we subtly combine the semantic structure information in the user–item bipartite graph with meta-paths in heterogeneous information networks (HIN), to filter some irrelevant items and obtain candidate items dynamically. By this means, the size of the discrete action space is effectively reduced from a new anger. The experimental results based on three benchmark datasets demonstrate the efficiency of our method with significant improvement over state-of-the-art. • Mining information in reviews and interaction data via the pretrained model. • Modeling long-term dynamic preferences of users accurately and discriminately. • Filtering irrelevant items and getting candidate items dynamically from a new anger. • Better interactive recommendation based on deep reinforcement learning. [ABSTRACT FROM AUTHOR]

We may not know the entire feature space in advance for real-world applications, and features can exist in a stream mode, called streaming features. Online streaming feature selection aims to select optimal streaming features on the fly and can be summarized into three main components: irrelevant feature discarding, relevant feature selecting, and redundant feature removing. Therefore, the core issue of the streaming feature selection framework is the calculation of the relationship between features. This paper applies Rough Set models to discover the feature relationships for the most crucial advantages: they do not require any domain knowledge and can measure the selected features as integral. After the formal definitions of feature relevance, irrelevance, and redundancy from the Rough Set perspective, we analyze and abstract the feature relationship calculation from three levels: Rough Set model, positive region, and consistency calculation. Then we design a novel general assembly Rough Set based Streaming Feature Selection Framework, named RS-SFSF, which could assemble new algorithms for different problems step by step. Researchers in different areas can quickly build the algorithms they need based on our new framework. To demonstrate the effectiveness of RS-SFSF, we derived four new algorithms based on RS-SFSF by using the classical Rough Set model, neighborhood Rough Set model, and fuzzy Rough Set model, respectively. Extensive experiments conducted on twelve real-world datasets indicate the efficiency of our new framework. • We summarize online streaming feature selection into three main components. • Formal definitions of feature relationships from the Rough Set perspective. • A novel general framework can assemble new algorithms for different problems. • Experiments on four new derived algorithms indicate the efficiency. [ABSTRACT FROM AUTHOR]

In various recommender systems, ratings and reviews are the main information to show user preferences. However, recommendation models that only use ratings, such as collaborative filtering, are vulnerable to data sparsity. And models only using review information will also suffer from the sparsity of reviews. On one hand, most ratings and reviews are interrelated and complementary, reviews may explain why a user gives a high or low rating to an item. On the other hand, ratings and reviews are numerical and textual information, respectively, and they reflect the preference of the user from a coarse-grained level and a fine-grained level A user may comment positively about some aspects of an item, even he gives a very low score to this item. There are specific information among each of them because of their heterogeneity. Therefore, it is possible to learn more accurate representation of users and items by effectively integrating ratings and text reviews from different views, that is, shared-view and specific-view. In this paper, we propose a Shared-view and Specific-view Information extraction model for Recommendation (SSIR), which integrates the information from reviews and interaction matrix to predict ratings Our model has two key components, including shared-view information extraction and specific-view exploitation. From the perspective of shared-view, SSIR jointly minimizes the loss of confusion adversarial and rating prediction loss to extract the shared information from reviews and user–item interaction matrix. For the specific-view part, SSIR applies orthogonal constraints on shared-view and specific-view modules to extract the discriminative features from reviews and interaction data. We fuse the features extracted from these two views to predict the final ratings. In addition, we use auxiliary reviews to deal with the sparsity problem of reviews. Experimental results on eight datasets show the effectiveness and robustness of our method, which could adapt to the recommendation scenarios with fewer reviews and ratings. • Mining information in reviews and interaction data from shared and specific views. • Utilizing the confusion adversarial loss to extract the shared features. • Enforcing orthogonal constraints to extract the specific features. • Shared-view information and specific-view information are synergized to recommend. [ABSTRACT FROM AUTHOR]

When extracting knowledge (or patterns) from multiple databases, the data from different databases might be too large in volume to be merged into one database for centralized mining on one computer, the local information sources might be hidden from a global decision maker due to privacy concerns, and different local databases may have different contribution to the global pattern. Dealing with multiple databases is essentially different from mining from a single database. In multi-database mining, the global patterns must be obtained by carefully analyzing the local patterns from individual databases. In this paper, we propose a nonlinear method, named KEMGP (kernel estimation for mining global patterns), to tackle this problem, which adopts kernel estimation to synthesizing local patterns for global patterns. We also adopt a method to divide all the data in different databases according to attribute dimensionality, which reduces the total space complexity. We test our algorithm on a customer management system, where the application is to obtain all globally interesting patterns by analyzing the individual databases. The experimental results show that our method is efficient. [Copyright &y& Elsevier]

Existing neural collaborative filtering (NCF) recommendation methods suffer from severe sparsity problem. Knowledge Graph (KG), which commonly consists of fruitful connected facts about items, presents an unprecedented opportunity to alleviate the sparsity problem. However, pure NCF models can hardly model the high-order connectivity in KG, and ignores complex pairwise correlations between user/item embedding dimensions. To address these problems, we propose a novel Knowledge graph enhanced Neural Collaborative Recommendation (K-NCR) framework, which effectively combines user–item interaction information and auxiliary knowledge information for recommendation task into three parts: (1) For items, the proposed propagating model learns the representation of item entity. It recursively aggregates information from its multi-hop neighbours in KG, and employs an attention mechanism to discriminate the importance of the relation type to mine users' potential preferences. (2) For users, another heterogeneous attention weights are leveraged to strengthen the embedding learning of users. (3) The user and item embeddings are then fed into a newly designed two-dimensional interaction map with convolutional hidden layers to model the complex pairwise correlations between their embedding dimensions explicitly. Extensive experimental results on three benchmark datasets demonstrate the effectiveness of our K-NCR framework. • An end-to-end knowledge graph enhanced neural recommendation is proposed. • An embedding propagation-based entity embedding framework is built. • An outer product based convolutional neural correlations framework is presented. [ABSTRACT FROM AUTHOR]