Improving Conversational Recommender Systems via Knowledge-Enhanced Temporal Embedding

EasyChair Preprint no. 10885

Abstract

Conversational recommender systems are becoming increasingly popular due to their potential to facilitate personalized interactions between users. However, one major challenge lies in accurately representing the semantic meaning of the conversational history to make relevant recommendations. In this paper, we propose a knowledge-enhanced model KITE to enhance conversational recommender systems. To achieve a more nuanced understanding of users' evolving interests and behaviors over time, a knowledge-enhanced temporal embedding is integrated into KITE to facilitate the encoding of temporal aspects into the representation of user dialogues. Our proposal is extensively evaluated on a real conversational dataset, and the experimental results demonstrate the effectiveness and superiority of our proposals in improving the accuracy and relevance of conversational recommender systems. Our work sheds light on the potential of leveraging advanced language models to enhance the performance of conversational recommender systems.

Keyphrases: Conversational Recommender Systems, Pre-trained Language Models, Temporal Embedding.

@Booklet{EasyChair:10885,
  author = {Chen Ji and Jilu Wang and Jie Xu and Wenxiao Liu and Zihong Yang and Feiran Huang and Chaozhuo Li},
  title = {Improving Conversational Recommender Systems via Knowledge-Enhanced Temporal Embedding},
  howpublished = {EasyChair Preprint no. 10885},

  year = {EasyChair, 2023}}