Improving preference elicitation in a conversational recommender system with active learning strategies

Conversational Recommender Systems are gaining more and more attention in the last years. They are characterized by the ability of establishing a multi-turn dialog with the user. Since those systems generally work in a cold-start situation, most of the conversation is devoted to the preference-elicitation step. However, in order to generate good recommendations, the user profile should be as rich as possible, which requires great user effort. In this paper, we investigate the application of Active Learning techniques for improving the preference elicitation step in a Conversational Recommender System. We compared different state-of-the-art techniques, and carried out a user study with 192 users in order to assess their effectiveness both in terms of recommendation accuracy and user effort. Results demonstrated that integrating item selection strategies based on item popularity improves the quality of the recommendations in terms of Hit Rate and nDCG, compared to a strategy based only on user-provided preferences.