In our complex modern world, we increasingly rely on automated systems to guide us in our decisionmaking. Intelligent recommender systems require knowledge about the users and about potential items of interest, but can also benefit from various sorts of background knowledge, e.g., how diferent human activities relate to one another. In this talk, I discuss several methods to bring together these various kinds of knowledge by means of knowledge graphs and neuro-symbolic explainable AI. Such methods draw on deep reinforcement learning or neural logic reasoning to provide explanations that allow users to better understand why particular items are being recommended. I will also discuss our work on how to mitigate bias and enable dialogue-based interaction in conversational recommender systems, along with datasets and code that we have released to promote further research on these topics.
1. Speaker