Sentimentally enhanced conversation recommender system

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