Sentimentally enhanced conversation recommender system

• DOI: 10.1007/s40747-024-01766-9
• Publication Date: 2025-01-08T07:12:57Z
• Authors: Fengjin Liu; Qiong Cao; Xianying Huang; Huaiyu Liu
• Keywords: Knowledge graph; Movie introduction; Sentiment analysis; Electronic computers. Computer science; QA75.5-76.95; Information technology; T58.5-58.64; Conversational recommender system

Conversation recommender system (CRS) aims to provide high-quality recommendations users in fewer conversation turns. Existing studies often rely on knowledge graphs enhance the representation of entity information. However, these methods tend overlook inherent incompleteness graphs, making it challenging for models fully capture users' true preferences. Additionally, they fail thoroughly explore emotional tendencies toward entities or effectively differentiate varying impacts different user Furthermore, responses generated by dialogue module are monotonous, lacking diversity and expressiveness, thus fall short meeting demands complex scenarios. To address shortcomings, we propose an innovative Sentimentally Enhanced Recommender System (SECR). First, construct a comprehensive highly optimized graph, termed MAKG, which provides rich complete set help model preferences more holistically. This significantly improves inference depth decision accuracy system. Second, deeply analyzing semantics dialogues, accurately identifies recommends those that best align with their refine recommendation strategy, design weighting mechanism quantify distinguish importance shaping Lastly, develop efficient text filter extract movie introductions from external data sources integrate them into dialogue, greatly enhancing semantic richness responses. Extensive experimental results two public CRS datasets demonstrate effectiveness our approach. Our code is released https://github.com/Janns0916/EECR .