Zhu Sun

Exploiting Implicit Item Relationships for Recommender Systems

Collaborative filtering inherently suffers from the data sparsity and cold start problems. Social networks have been shown useful to help alleviate these issues. However, social connections may not be available in many real systems, whereas implicit item relationships are lack of study. In this paper, we propose a novel matrix factorization model by taking into account implicit item relationships. Specifically, we employ an adapted association rule technique to reveal implicit item relationships in terms of item-to-item and group-to-item associations, which are then used to regularize the generation of low-rank user- and item-feature matrices. Experimental results on four real-world datasets demonstrate the superiority of our proposed approach against other counterparts.

Learning Hierarchical Feature Influence for Recommendation by Recursive Regularization

Existing feature-based recommendation methods incorporate auxiliary features about users and/or items to address data sparsity and cold start issues. They mainly consider features that are organized in a flat structure, where features are independent and in a same level. However, auxiliary features are often organized in rich knowledge structures (e.g. hierarchy) to describe their relationships. In this paper, we propose a novel matrix factorization framework with recursive regularization -- ReMF, which jointly models and learns the influence of hierarchically-organized features on user-item interactions, thus to improve recommendation accuracy. It also provides characterization of how different features in the hierarchy co-influence the modeling of user-item interactions. Empirical results on real-world data sets demonstrate that ReMF consistently outperforms state-of-the-art feature-based recommendation methods.

Interacting Attention-gated Recurrent Networks for Recommendation

Capturing the temporal dynamics of user preferences over items is important for recommendation. Existing methods mainly assume that all time steps in user-item interaction history are equally relevant to recommendation, which however does not apply in real-world scenarios where user-item interactions can often happen accidentally. More importantly, they learn user and item dynamics separately, thus failing to capture their joint effects on user-item interactions. To better model user and item dynamics, we present the Interacting Attention-gated Recurrent Network (IARN) which adopts the attention model to measure the relevance of each time step. In particular, we propose a novel attention scheme to learn the attention scores of user and item history in an interacting way, thus to account for the dependencies between user and item dynamics in shaping user-item interactions. By doing so, IARN can selectively memorize different time steps of a users history when predicting her preferences over different items. Our model can therefore provide meaningful interpretations for recommendation results, which could be further enhanced by auxiliary features. Extensive validation on real-world datasets shows that IARN consistently outperforms state-of-the-art methods.

MRLR: Multi-level Representation Learning for Personalized Ranking in Recommendation.

Representation learning (RL) has recently proven to be effective in capturing local item relationships by modeling item co-occurrence in individual users interaction record. However, the value of RL for recommendation has not reached the full potential due to two major drawbacks: 1) recommendation is modeled as a rating prediction problem but should essentially be a personalized ranking one; 2) multi-level organizations of items are neglected for fine-grained item relationships. We design a unified Bayesian framework MRLR to learn user and item embeddings from a multi-level item organization, thus benefiting from RL as well as achieving the goal of personalized ranking. Extensive validation on real-world datasets shows that MRLR consistently outperforms state-of-the-art algorithms.

BPRH: Bayesian Personalized Ranking for Heterogeneous Implicit Feedback

Abstract Personalized recommendation for online service systems aims to predict potential demand by analysing user preference. User preference can be inferred from heterogeneous implicit feedback (i.e. various user actions) especially when explicit feedback (i.e. ratings) is not available. However, most methods either merely focus on homogeneous implicit feedback (i.e. target action), e.g., purchase in shopping websites and forward in Twitter, or dispose heterogeneous implicit feedback without the investigation of its speciality. In this paper, we adopt two typical actions in online service systems, i.e., view and like, as auxiliary feedback to enhance recommendation performance, whereby we propose a Bayesian personalized ranking method for heterogeneous implicit feedback (BPRH). Specifically, items are first classified into different types according to the actions they received. Then by analysing the co-occurrence of different types of actions, which is one of the fundamental speciality of heterogeneous implicit feedback systems, we quantify their correlations, based on which the difference of users’ preference among different types of items is investigated. An adaptive sampling strategy is also proposed to tackle the unbalanced correlation among different actions. Extensive experimentation on three real-world datasets demonstrates that our approach significantly outperforms state-of-the-art algorithms.

Effective Recommendation with Category Hierarchy

Although flat item category structure where categories are independent in a same level has been well studied to enhance recommendation performance, in many real applications, item category is often organized in hierarchies to reflect the inherent correlations among categories. In this paper, we propose a novel matrix factorization model by exploiting category hierarchy from the perspectives of users and items for effective recommendation. Specifically, a user (an item) can be influenced (characterized) by her preferred categories (the categories it belongs to) in the hierarchy. We incorporate how different categories in the hierarchy co-influence a user and an item. Empirical results show the superiority of our approach against other counterparts.

Recurrent knowledge graph embedding for effective recommendation

Knowledge graphs (KGs) have proven to be effective to improve recommendation. Existing methods mainly rely on hand-engineered features from KGs (e.g., meta paths), which requires domain knowledge. This paper presents RKGE, a KG embedding approach that automatically learns semantic representations of both entities and paths between entities for characterizing user preferences towards items. Specifically, RKGE employs a novel recurrent network architecture that contains a batch of recurrent networks to model the semantics of paths linking a same entity pair, which are seamlessly fused into recommendation. It further employs a pooling operator to discriminate the saliency of different paths in characterizing user preferences towards items. Extensive validation on real-world datasets shows the superiority of RKGE against state-of-the-art methods. Furthermore, we show that RKGE provides meaningful explanations for recommendation results.

Learning hierarchical category influence on both users and items for effective recommendation

Item category has proven to be useful additional information to address the data sparsity and cold start problems in recommender systems. Although categories have been well studied in which they are independent and structured in a flat form, in many real applications, item category is often organized in a richer knowledge structure - category hierarchy, to reflect the inherent correlations among different categories. In this paper, we propose a novel latent factor model by exploiting category hierarchy from the perspectives of both users and items for effective recommendation. Specifically, a user can be influenced by her preferred categories in the hierarchy. Similarly, an item can be characterized by the associated categories in the hierarchy. We incorporate the influence that different categories have towards a user and an item in the hierarchical structure. Experimental results on two real-world data sets demonstrate that our method consistently outperforms the state-of-the-art category-aware recommendation algorithms.

Online Reputation Fraud Campaign Detection in User Ratings

Reputation fraud campaigns (RFCs) distort the reputations of rated items, by generating fake ratings through multiple spammers. One effective way of detecting RFCs is to characterize their collective behaviors based on rating histories. However, these campaigns are constantly evolving and changing tactics to evade detection. For example, they can launch early attacks on the items to quickly dominate the reputations. They can also whitewash themselves through creating new accounts for subsequent attacks. It is thus challenging for existing approaches working on historical data to promptly react to such emerging fraud activities. In this paper, we conduct RFC detection in online fashion, so as to spot campaign activities as early as possible. This leads to a unified and scalable optimization framework, FRAUDSCAN, that can adapt to emerging fraud patterns over time. Empirical analysis on two real-world datasets validates the effectiveness and efficiency of the proposed framework.

A Unified Latent Factor Model for Effective Category-Aware Recommendation

Our data analysis on real-world datasets shows that user preferences are intimately related with item categories, implying the non-negligible of category information for effective recommendation. Thus, in this paper, step by step we propose a unified item-category latent factor model by considering user-category, item-category and category-category interactions. Our approach can be applied to both the situations where an item belongs to either a single category (one-to-one) or multiple categories (one-to-many). Finally, empirical studies on the real-world datasets demonstrate the superiority of our approach in comparison with other counterparts.