Meta-CRS: A Dynamic Meta-Learning Approach for Effective Conversational Recommender System

被引:3
作者
Ni, Yuxin [1 ]
Xia, Yunwen [2 ]
Fang, Hui [3 ,4 ]
Long, Chong [5 ]
Kong, Xinyu [5 ]
Li, Daqian [5 ]
Yang, Dong [5 ]
Zhang, Jie [2 ]
机构
[1] Nanyang Technol Univ, 50 Nanyang Ave, Singapore 639798, Singapore
[2] Nanyang Technol Univ, Sch Comp Sci & Engn, 50 Nanyang Ave, Singapore 639798, Singapore
[3] Shanghai Univ Finance & Econ, RIIS, 100 Wudong Rd, Shanghai 200433, Peoples R China
[4] Shanghai Univ Finance & Econ, SIME, 100 Wudong Rd, Shanghai 200433, Peoples R China
[5] Ant Grp, Z Space 556 Xixi Rd, Hangzhou, Peoples R China
基金
上海市自然科学基金; 中国国家自然科学基金;
关键词
Conversational recommender system; reinforcement learning; meta learning; prior knowledge; knowledge graph; dynamic graph;
D O I
10.1145/3604804
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Conversational recommender system (CRS) enhances the recommender system by acquiring the latest user preference through dialogues, where an agent needs to decide "whether to ask or recommend", "which attributes to ask", and "which items to recommend" in each round. To explore these questions, reinforcement learning is adopted in most CRS frameworks. However, existing studies somewhat ignore to consider the connection between the previous rounds and the current round of the conversation, which might lead to the lack of prior knowledge and inaccurate decisions. In this view, we propose to facilitate the connections between different rounds of conversations in a dialogue session through deep transformer-based multi-channel meta-reinforcement learning, so that the CRS agent can decide each action/decision based on previous states, actions, and their rewards. Besides, to better utilize a user's historical preferences, we propose a more dynamic and personalized graph structure to support the conversation module and the recommendationmodule. Experiment results on five real-world datasets and an online evaluation with real users in an industrial environment validate the improvement of our method over the state-of-the-art approaches and the effectiveness of our designs.
引用
收藏
页数:27
相关论文
共 62 条
[1]  
[Anonymous], 2016, P ICLR
[2]  
Antoniou Antreas, 2019, 7 INT C LEARN REPR I, DOI DOI 10.1145/3351556.3351574
[3]  
Beck J, 2024, Arxiv, DOI [arXiv:2301.08028, 10.48550/ARXIV.2301.08028]
[4]  
Bordes A., 2013, Adv. Neural Inf. Process. Syst., V26
[5]  
Chen QB, 2019, 2019 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING AND THE 9TH INTERNATIONAL JOINT CONFERENCE ON NATURAL LANGUAGE PROCESSING (EMNLP-IJCNLP 2019), P1803
[6]   Q&R: A Two-Stage Approach toward Interactive Recommendation [J].
Christakopoulou, Konstantina ;
Beutel, Alex ;
Li, Rui ;
Jain, Sagar ;
Chi, Ed H. .
KDD'18: PROCEEDINGS OF THE 24TH ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY & DATA MINING, 2018, :139-147
[7]   Towards Conversational Recommender Systems [J].
Christakopoulou, Konstantina ;
Radlinski, Filip ;
Hofmann, Katja .
KDD'16: PROCEEDINGS OF THE 22ND ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING, 2016, :815-824
[8]   Meta Policy Learning for Cold-Start Conversational Recommendation [J].
Chu, Zhendong ;
Wang, Hongning ;
Xiao, Yun ;
Long, Bo ;
Wu, Lingfei .
PROCEEDINGS OF THE SIXTEENTH ACM INTERNATIONAL CONFERENCE ON WEB SEARCH AND DATA MINING, WSDM 2023, VOL 1, 2023, :222-230
[9]   Leveraging Long Short-Term User Preference in Conversational Recommendation via Multi-agent Reinforcement Learning [J].
Deng, Yang ;
Li, Yaliang ;
Ding, Bolin ;
Lam, Wai .
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 2023, 35 (11) :11541-11555
[10]   Unified Conversational Recommendation Policy Learning via Graph-based Reinforcement Learning [J].
Deng, Yang ;
Li, Yaliang ;
Sun, Fei ;
Ding, Bolin ;
Lam, Wai .
SIGIR '21 - PROCEEDINGS OF THE 44TH INTERNATIONAL ACM SIGIR CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION RETRIEVAL, 2021, :1431-1441