Leveraging Historical Interaction Data for Improving Conversational Recommender System

Recently, conversational recommender system (CRS) has become With the rapid development of intelligent agents in e-commerce an emerging and practical research topic. Most of the existing CRS platforms, conversational recommender system (CRS) [5, 6, 9] has methods focus on learning effective preference representations for become an emerging research topic in seeking to provide highquality users from conversation data alone. While, we take a new perspective recommendations to users through conversations. Generally, to leverage historical interaction data for improving CRS. a CRS consists of a conversation module and a recommendation For this purpose, we propose a novel pre-training approach to module. The conversation module focuses on acquiring users' preference integrating both item-based preference sequence (from historical via multi-turn interaction, and the recommendation module interaction data) and attribute-based preference sequence (from conversation focuses on how to utilize the inferred preference information to data) via pre-training methods. We carefully design two recommend suitable items for users.