Exploiting the Structure of User Feedback in Recommender Systems 424 views

From shopping to dining, people consume items on digital service platforms as a part of their daily life routine. In the meanwhile, people leave feedback about the consumptions, e.g., browsing products or writing reviews. By leveraging user feedback, recommender systems predict the items that users would be interested in and then make recommendations. Recommender systems are important for both users and service providers. They not only save users' efforts on finding relevant items but also achieve the providers' objectives on improving user satisfaction. In this thesis, we study the problems of making contextualized recommendations and providing explanations for recommendations, since these are highly desired properties of recommender systems. Modeling the dependence among user feedback is essential to address these two problems. Various research efforts have been devoted to improving the recommendation quality and the quality of explanations for recommendations. Nevertheless, the structure among user feedback is overlooked. In the thesis, we argue that explicitly exploiting the structure (e.g., sequential structure, or graph structure) among user feedback allows us to effectively recognize the dependence among feedback, consequently improving the recommendation quality and the quality of explanations for recommendations. First, for contextualized recommendations, we developed solutions that exploit the sequential structure among user feedback in regard to the temporal information and category information of user feedback. Second, to provide explanations for recommendations, we developed a solution that exploits the graph structure among user feedback. We have demonstrated the effectiveness of our solutions through experiments on large datasets. This thesis shows that exploiting structure among feedback helps build an effective and explainable recommender system.

PHD (Doctor of Philosophy)

recommender systems; structural information; effectiveness and explainability

English

2021-08-04