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

TL;DR

COMPASS addresses the explainability gap in conversational recommender systems by integrating knowledge graphs with large language models to reason over user preferences. It introduces graph entity captioning to bridge KG-LLM modality gaps and knowledge-aware instruction tuning to enable cross-modal reasoning, producing textual, interpretable preference summaries. As a plug-in, COMPASS improves recommendation performance and explainability across diverse CRS models, validated on ReDial and INSPIRED with significant gains over baselines and thorough ablations. The work demonstrates the practical potential of KG-augmented LLMs for transparent, knowledge-rich conversational recommendations.

Abstract

Conversational Recommender Systems (CRSs) aim to provide personalized recommendations by capturing user preferences through interactive dialogues. Explainability in CRSs is crucial as it enables users to understand the reasoning behind recommendations, increasing system transparency and trustworthiness. However, current CRSs often leverage knowledge graphs (KGs) or language models to extract and represent user preferences as latent vectors, which limits their explainability. Large language models (LLMs) offer powerful reasoning capabilities that can bridge this gap by generating human-understandable preference summaries. However, effectively reasoning over user preferences in CRSs remains challenging as LLMs pre-trained on large-scale corpora may not be well-suited for analyzing user preferences. While KGs provide rich domain knowledge, integrating them with LLMs encounters a significant modality gap between structured KG information and unstructured conversations. In this paper, we propose COMPASS, a plug-and-play framework that synergizes LLMs and KGs to reason over user preferences, enhancing the performance and explainability of existing CRSs. COMPASS employs a two-stage training approach: first, it bridges the gap between the structured KG and natural language through novel graph entity captioning pre-training. Next, COMPASS optimizes user preference reasoning via knowledge-aware instruction fine-tuning, where the LLM learns to reason and summarize user preferences from dialogue histories and KG-augmented context. This enables COMPASS to perform knowledge-aware reasoning and generate interpretable user preferences that can seamlessly integrate with existing CRS models for improving recommendation performance and explainability. Our experiments on benchmark datasets demonstrate the effectiveness of COMPASS in improving various CRS models.

Figures (4)

• Figure 1: Conventional CRSs lack explainability in user preference modeling by extracting hidden representation, while our approach enhances transparency by generating interpretable user preference in text.
• Figure 2: The overall framework of our COMPASS. COMPASS consists of three components: the graph encoder, the adapter, and the LLM. The adapter aligns the knowledge graph to the LLM. COMPASS follows a two-stage training paradigm - (a) Graph Entity Captioning and (b) Knowledge-aware Instruction Tuning. Once trained, COMPASS can be easily integrated with existing CRS models for (c) Preference Enhanced Recommendation.
• Figure 3: Comparison of recommendation performance on the INSPIRED dataset with different enhancers. Green percentages show improvements over baselines.
• Figure 4: LLM-simulated user rankings for User Preference Alignment and Explainability across ReDial and INSPIRED datasets.