Combining Multiple Information Sources And Incorporating Implicit Feedback For Recommender Systems

For all the benefits of the information abundance and communication technology, the "information overload" is one of the digital-age dilemmas we are confronted with. Recommender systems (RSs) are instrumental in tackling this problem. They help of-fer potential interesting information to individual consumers and allow online users to quickly find the personalized information that fits their needs. RSs are nowadays ubiq-uitous in various domains and e-commerce platforms. They are used to recommend books, point of interest and activities, music, movie, and references.Collaborative filtering (CF) approaches play a central role in traditional recom-mender systems, which are extensively investigated in research community and widely used in industry. They are based on the simple intuition that if users rated items sim-ilarly in the past, then they will be likely to rate other items similarly in the future. CF models, however, suffer from data sparsity and the imbalance of ratings. They perform poorly on cold users and cold items for which there are no or few data. To overcome these weaknesses, additional sources of information are integrated into RSs. One research thread, which we call topic matrix factorization (Topic MF), is to inte-grate ratings with item contents or reviews text. Another research thread, which we call social matrix factorization (Social MF), is to combine ratings with social relations. N-evertheless, both Social MF and Topic MF ignore some useful information, either item reviews or social relations. Contrast to integrating more data sources into RSs, another way is to mine the limited data information more deeply; for example, factorizing only the rating matrix and exploiting the implicit feedback from it. Nevertheless, there is no related work on how to merge the two ways of integrating auxiliary sources and of mining implicit feedback.This thesis focuses on combining multiple information sources and incorporating implicit feedback for recommender systems. The main contributions of the dissertation are outlined as follows.1. Proposing a novel synthetic approach MR3 to jointly model user-item ratings, social network structure, and item reviews for rating prediction; and along with an extended Social MF method which exploits the ratings and social relations more tightly by capturing the graph structure of neighbors.2. Extending the proposed approach to obtain a new model MR3++ by incorpo-rating implicit feedback from ratings to enhance its capability and to demonstrate its flexibility; the extension model mines the limited information more deeply by intro-ducing implicit features which captures the influence of rated items.We demonstrate our methods empirically on two real-life datasets over various state-of-the-art algorithms. Further we analyse the impact of their components and the sensitivity of their hyper-parameters.