HyCoRec: Hypergraph-Enhanced Multi-Preference Learning for Alleviating Matthew Effect in Conversational Recommendation

Conversational Recommender Systems (CRSs) often suffer from the Matthew effect, where popular items receive disproportionate exposure while niche items are ignored. While this phenomenon is well-documented in static settings, it is exacerbated in dynamic conversational environments where user-system feedback loops continuously reinforce popularity bias. This paper investigates how to alleviate this effect by better capturing diverse user preferences during ongoing dialogues.

The authors propose HyCoRec (Hypergraph-Enhanced Multi-Preference Learning for Alleviating Matthew Effect in Conversational Recommendation). The framework operates through two main components:

1. Hypergraph-Enhanced Multi-Preference Learning: Instead of relying on simple pairwise relationships, the model constructs multiple hypergraphs (item-based, entity-based, and word-based) to capture complex, high-order relationships between items, entities, and keywords. It further incorporates review-based and knowledge-based preferences to build a comprehensive profile of the user.
2. Hypergraph-Aware CRS: This component integrates the learned multi-aspect preferences into both the recommendation task (predicting items) and the conversational task (generating natural language responses). By fusing these diverse preference signals, the system aims to provide more personalized and diverse recommendations that resist the "rich-get-richer" cycle.

Experimental results on two benchmark datasets demonstrate that HyCoRec achieves state-of-the-art performance in both recommendation accuracy and response generation. The authors show that by explicitly modeling multi-aspect preferences via hypergraphs, the system effectively reduces the popularity bias inherent in standard conversational recommenders, thereby mitigating the Matthew effect as interactions progress over time.

AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.