Adapting User Preference to Online Feedback in Multi-round Conversational Recommendation

This paper concerns the estimation of user preference on attributes and items in multi-round conversational recommender systems (CRS).Compared to the static recommender systems and the single-round CRS, a multi-round CRS interacts with user by asking questions about attributes and recommending items multiple times in one conversation. Multi-round CRS such as EAR have been proposed in which the user's online feedback at both attribute level and item level can be utilized to predict user preference and make recommendation. Though preliminary success has been shown, existing user preference models in CRS (e.g., the Factorization Machine in EAR) usually use the online feedback information as independent features or the training instances, overlooking the hierarchy relation between attribute-level and item-level feedback signals. The hierarchy relation can be used to, for example, more precisely identify the reasons (e.g. some certain attributes) that trigger the user rejection to an item, leading to more fine-grained utilization of the feedback information. This paper proposes a novel preference prediction model tailored for multi-round CRS, called Feedback-guided Preference Adaptation Network (FPAN). In FPAN, two gating modules are designed to respectively adapt the original user embedding and item-level feedback, both according to the online attribute-level feedback. By considering the hierarchy relation, the gating modules utilize the fine-grained attribute-level feedback to revise the user embedding and coarse-grained item-level feedback, achieving more accurate user preference prediction. Experimental results on two benchmarks showed that FPAN outperformed the state-of-the-art user preference models adopted in current CRS, and the whole multi-round CRS can also be enhanced by using FPAN as its recommender component.