Comprehending Knowledge Graphs with Large Language Models for Recommender Systems

TL;DR

This work tackles key limitations of KG-enhanced recommender systems, notably missing facts, semantic loss from ID-based KG representations, and poor capture of high-order relations. It introduces CoLaKG, a two-stage framework that uses LLMs to comprehend local item-centered KG subgraphs and to retrieve globally related items, followed by retrieval-augmented representation learning and cross-modal alignment with ID embeddings. The approach yields consistent improvements over a wide range of baselines across four real-world datasets and demonstrates robustness to sparsity and missing KG facts, while keeping online inference efficient. The method's ability to inject rich semantic signals from KGs into recommendations offers a practical path toward more accurate and Knowledge-grounded personalization.

Abstract

In recent years, the introduction of knowledge graphs (KGs) has significantly advanced recommender systems by facilitating the discovery of potential associations between items. However, existing methods still face several limitations. First, most KGs suffer from missing facts or limited scopes. Second, existing methods convert textual information in KGs into IDs, resulting in the loss of natural semantic connections between different items. Third, existing methods struggle to capture high-order connections in the global KG. To address these limitations, we propose a novel method called CoLaKG, which leverages large language models (LLMs) to improve KG-based recommendations. The extensive knowledge and remarkable reasoning capabilities of LLMs enable our method to supplement missing facts in KGs, and their powerful text understanding abilities allow for better utilization of semantic information. Specifically, CoLaKG extracts useful information from KGs at both local and global levels. By employing the item-centered subgraph extraction and prompt engineering, it can accurately understand the local information. In addition, through the semantic-based retrieval module, each item is enriched by related items from the entire knowledge graph, effectively harnessing global information. Furthermore, the local and global information are effectively integrated into the recommendation model through a representation fusion module and a retrieval-augmented representation learning module, respectively. Extensive experiments on four real-world datasets demonstrate the superiority of our method.

Figures (7)

• Figure 1: An illustrative diagram demonstrating the potential issues of existing KG-based recommendation methods.
• Figure 2: The framework of our proposed CoLaKG.
• Figure 3: The prompt template for Local KG comprehension.
• Figure 4: Hyperparameter study of the number of retrieved neighbors ($k$) and sampled number of 2-hop items within the prompt ($m$) on four datasets.
• Figure 5: Performance comparison on different user groups, where a smaller group ID indicates fewer interaction records.