Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

Reasoning over User Preferences: Knowledge Graph-Augmented LLMs for Explainable Conversational Recommendations

Created by
  • Haebom

Author

Zhangchi Qiu, Linhao Luo, Shirui Pan, Alan Wee-Chung Liew

Outline

To improve the performance and explainability of conversational recommender systems (CRS), we propose the COMPASS framework, which combines LLM and knowledge graphs (KGs). COMPASS pre-trains graph entity captioning to bridge the gap between KG information and natural language, and optimizes user preference inference through knowledge-based command fine-tuning. This allows COMPASS to generate interpretable user preferences and integrate them into existing CRS models, improving recommendation performance and explainability.

Takeaways, Limitations

•
Takeaways:
◦
A novel framework that integrates LLM and KG to improve the explainability and performance of CRS.
◦
Bridging the Gap Between KG and Natural Language with Graph Entity Captioning Pretraining
◦
Enhancing LLM's ability to infer user preferences through knowledge-based command fine-tuning.
◦
Generate interpretable user preferences that can be easily integrated into existing CRS models.
◦
Demonstrating improved performance of various CRS models on benchmark datasets.
•
Limitations:
◦
Only experimental results for specific datasets and models are presented, requiring further review of generalizability.
◦
Computational cost and training time need to be considered.
◦
May exhibit performance dependent on KG quality
👍
Made with Slashpage