Enhancing Conversational Recommender Systems with Tree-Structured Knowledge and Pretrained Language Models

Yov Ren, Chao Wang, Chuan Qin, Dazhong Shen, Hui Xiong

Keywords: Conversational Recommendation Systems, Knowledge Graph, Pretrained Language Model

Abstract:

Conversational recommender systems (CRS) have emerged as a key enhancement to traditional recommendation systems, offering interactive and explainable recommendations through natural dialogue.
Recent advancements in pretrained language models (PLMs) have significantly improved the conversational capabilities of CRS, enabling more fluent and context-aware interactions.
However, PLMs still face challenges, including hallucinations—where the generated content can be factually inaccurate—and difficulties in providing precise, entity-specific recommendations.
To address these challenges, we propose the PCRS-TKA framework, which integrates PLMs with knowledge graphs (KGs) through prompt-based learning. By incorporating tree-structured knowledge from KGs, our framework grounds the PLM in factual information, thereby enhancing the accuracy and reliability of the recommendations. Additionally, we design a user preference extraction module to improve the personalization of recommendations and introduce an alignment module to ensure semantic consistency between dialogue text and KG data. Extensive experiments demonstrate that PCRS-TKA outperforms existing methods in both recommendation accuracy and conversational fluency.

Primary Area: other topics in machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9496