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

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