LLM-Guided Multi-View Hypergraph Learning for Human-Centric Explainable Recommendation

TL;DR

This work tackles the challenge of modeling multifaceted human preferences in recommendations by integrating large language models (LLMs) with multi-view hypergraph learning. The core approach, LLMHG, first extracts Interest Angles (IAs) from user histories using LLMs, then builds a multi-view hypergraph where each view corresponds to an IA and items are grouped into angle-specific hyperedges. To overcome LLM limitations, the framework applies hypergraph structure learning to refine prototypes and weights, leveraging intra-edge and inter-edge cues, plus a normalization via the hypergraph Laplacian, before fusing with sequential embeddings for prediction. Empirically, LLMHG improves over strong baselines on ML-1M, Amazon Beauty, and Amazon Toys, with GPT-4 generally delivering the best results, albeit at higher computational cost; the work highlights a promising plug-and-play path to human-centric, explainable recommendations that leverage both world knowledge and high-order relational structure.

Abstract

As personalized recommendation systems become vital in the age of information overload, traditional methods relying solely on historical user interactions often fail to fully capture the multifaceted nature of human interests. To enable more human-centric modeling of user preferences, this work proposes a novel explainable recommendation framework, i.e., LLMHG, synergizing the reasoning capabilities of large language models (LLMs) and the structural advantages of hypergraph neural networks. By effectively profiling and interpreting the nuances of individual user interests, our framework pioneers enhancements to recommendation systems with increased explainability. We validate that explicitly accounting for the intricacies of human preferences allows our human-centric and explainable LLMHG approach to consistently outperform conventional models across diverse real-world datasets. The proposed plug-and-play enhancement framework delivers immediate gains in recommendation performance while offering a pathway to apply advanced LLMs for better capturing the complexity of human interests across machine learning applications.

Figures (5)

• Figure 1: LLMHG includes four major steps: interest angle extraction, construction of a multi-view hypergraph centered on interest angles, hypergraph structure learning for LLM content refinement, and representation fusion for recommendation prediction.
• Figure 2: The real case studies (ML-1M) on our (c) LLMHG and ablation models, i.e., (a) LLMHG w/o interest angle generation and (b) LLMHG w/o hypergraph structure learning .
• Figure 3: Sensitivity analysis of sequence truncation length $l_{tru}$ and intra and inter structure learning coefficient $\beta$ on HR and NDCG performance based on ML-1M and Amazon Beauty benchmarks.
• Figure 4: The analysis between prototype correction weight $\lambda$ and sequence truncation length $l_{tru}$.
• Figure 5: The prompt examples and real cases for LLMHG.