G-Refer: Graph Retrieval-Augmented Large Language Model for Explainable Recommendation

TL;DR

G-Refer addresses explainable recommendations by integrating graph retrieval with large language models. It uses a hybrid graph retriever to extract explicit structural and semantic CF signals, translates graph knowledge into natural language, and applies retrieval-augmented fine-tuning with LoRA (RAFT) to improve explanation generation. A knowledge-pruning step reduces noisy data, improving training efficiency and explanation quality. Across three public datasets, G-Refer outperforms state-of-the-art baselines in semantic explainability and stability, with human evaluators favoring its explanations, demonstrating the practical value of explicit CF signals and graph-to-text translation in explainable recommendation.

Abstract

Explainable recommendation has demonstrated significant advantages in informing users about the logic behind recommendations, thereby increasing system transparency, effectiveness, and trustworthiness. To provide personalized and interpretable explanations, existing works often combine the generation capabilities of large language models (LLMs) with collaborative filtering (CF) information. CF information extracted from the user-item interaction graph captures the user behaviors and preferences, which is crucial for providing informative explanations. However, due to the complexity of graph structure, effectively extracting the CF information from graphs still remains a challenge. Moreover, existing methods often struggle with the integration of extracted CF information with LLMs due to its implicit representation and the modality gap between graph structures and natural language explanations. To address these challenges, we propose G-Refer, a framework using graph retrieval-augmented large language models (LLMs) for explainable recommendation. Specifically, we first employ a hybrid graph retrieval mechanism to retrieve explicit CF signals from both structural and semantic perspectives. The retrieved CF information is explicitly formulated as human-understandable text by the proposed graph translation and accounts for the explanations generated by LLMs. To bridge the modality gap, we introduce knowledge pruning and retrieval-augmented fine-tuning to enhance the ability of LLMs to process and utilize the retrieved CF information to generate explanations. Extensive experiments show that G-Refer achieves superior performance compared with existing methods in both explainability and stability. Codes and data are available at https://github.com/Yuhan1i/G-Refer.

Figures (5)

• Figure 1: Our proposed pipeline G-Refer facilitates explainable recommendation with three key components: (1) Hybrid Graph Retrieval employs multi-granularity retrievers to retrieve explicit CF signals and formulated as human-readable text by the Graph Translation; (2) Knowledge Pruning eliminates noise and improves training efficiency; and (3) Retrieval-augmented Fine-tuning instructs LLMs to leverage retrieved CF information in generating informative explanation.
• Figure 2: The illustration of path-level retriever.
• Figure 3: Human evaluation comparing XRec and G-Refer.
• Figure 4: Performance of different retrieved number $k$.
• Figure 5: Efficiency Analysis of G-Refer.