#41 NORA/NAIL Webinar: Toward More Natural Interactions with Conversational Recommender Systems

On Friday June 11, Krisztian Balog joined us to give a talk about conversational recommender systems. Krisztian is a professor at the University of Stavanger.

Recommender systems have become indispensable tools that help users navigate large collections to find items of potential interest. The conversational paradigm naturally lends itself to this problem, as it allows for rich interactions with the system using natural language. During a dialogue, the system can, among others, elicit preferences, provide explanations for the suggestions, and receive feedback on the recommendations. In this talk, I will address some key modeling and evaluation questions that reflect the unique challenges of the conversational setting. First, I will present an approach for generating natural language summaries of user preferences, which can be explicitly scrutinized, thereby giving users more direct control over the recommendations they receive. Second, I will address how to robustly interpret natural language refinements (or critiques) with respect to a given anchor item. Third, I will discuss the use of user simulation as an alternative means of end-to-end system evaluation.

Time: June 11, 2021 2:00 PM–3:00 PM