Sigir2022 𞓜 user preference modeling in conversational recommendation system

Thesis link ：https://irlab.science.uva.nl/wp-content/papercite-data/pdf/ren-2022-variational.pdf

This paper proposes a dialogue recommendation system based on user preferences （User Preferences for Conversational Recommendation,UPCR）, Infer users' long-term and short-term preferences from historical dialogues and current dialogues respectively , To solve the problem of no marking , We regard long-term and short-term user preferences as two independent implicit variables , Variable Bayesian inference is used to approximate the exact posterior probability distribution , At the recommendation stage , In addition to user preferences , We also introduce external knowledge to enhance the relevance between topics and objects , Experimental results show the effectiveness of the proposed method .

The contributions are summarized as follows ：

1. In the topic based conversational recommendation system , We proposed UPCR To mine users' long-term preferences and short-term preferences .

2. To solve the problem of user preference without annotation , We model user preferences , The variable Bayesian algorithm is used to infer users' long-term preferences and short-term preferences .

3. In addition to user preferences , We also introduce external knowledge to improve the accuracy of topic prediction and recommendation .

4.UPCR The effect on two conversational recommendation datasets exceeds that of other models , Proved UPCR The effectiveness of the .