Recognizing users feedback from non-verbal communicative acts in conversational recommender systems

Conversational recommender systems produce personalized recommendations of potentially useful items by utilizing natural language dialogues for detecting user preferences, as well as for providing recommendations. In this work we investigate the role of affective factors such as attitudes, emotions, likes and dislikes in conversational recommender systems and how they can be used as implicit feedback to improve the information filtering process. We thus developed a multimodal framework for recognizing the attitude of the user during their conversation with DIVA, a Dress-shopping InteractiVe Assistant aimed at recommending fashion apparel. Wee took into account speech prosody, body poses and facial expressions for providing implicit feedback to the system and for refining the recommendation accordingly. The shopping assistant has been embodied in the Social Robot NAO and has been tested in the dress shopping scenario. Our experimental results show that the proposed method is a promising way to implicitly profile the user and improve the performance of recommendations when explicit feedback is not available, thus demonstrating its effectiveness and viability.