Learning from user interactions for recommending queries and items

People use search engines and recommender systems for multiple purposes in daily life – they may search for some specific information with queries, or they may just want to watch movies and do online shopping for entertainment. In many cases it is not clear what exactly users want: what information they are looking for or what items they are interested in. Thus, search engines and recommender systems need to become pro-active and provide users with auxiliary information so as to gain a better understanding of users' goals. E.g., search engines could suggest queries and recommender systems could offer lists of options that may address the users’ needs. In this thesis, we investigate how to learn from user interaction data and thus make satisfactory recommendations for both search engines and recommender systems. We propose an attention-based hierarchical neural architecture for query suggestion, a joint neural collaborative filtering approach for general recommendation, a dynamic co-attention network model for session-based recommendation, and an intent-aware end-to-end neural approach for diversified sequential recommendation. We also conduct comprehensive experiments and analysis to verify the effectiveness of our proposals. We finally list several possible directions for future work based on the research in this thesis.