MixRec: Heterogeneous Graph Collaborative Filtering

TL;DR

The paper addresses behavior heterogeneity and intent diversity in collaborative filtering by learning disentangled latent intents across multiple interaction types, represented as a tensor $\\mathcal{X} \\in \\mathbb{R}^{I\\times J\\times K}$. It introduces MixRec, which combines a parameterized heterogeneous hypergraph for intent disentanglement with relation-aware multiplex graph learning and a hierarchical, adaptive contrastive augmentation framework. Empirical results on three public datasets show MixRec consistently outperforms state-of-the-art baselines, with ablations confirming the importance of intent disentanglement and adaptive augmentation. The approach improves robustness under data sparsity and remains efficient due to a lightweight hypergraph GNN, with code released for public use.

Abstract

For modern recommender systems, the use of low-dimensional latent representations to embed users and items based on their observed interactions has become commonplace. However, many existing recommendation models are primarily designed for coarse-grained and homogeneous interactions, which limits their effectiveness in two critical dimensions. Firstly, these models fail to leverage the relational dependencies that exist across different types of user behaviors, such as page views, collects, comments, and purchases. Secondly, they struggle to capture the fine-grained latent factors that drive user interaction patterns. To address these limitations, we present a heterogeneous graph collaborative filtering model MixRec that excels at disentangling users' multi-behavior interaction patterns and uncovering the latent intent factors behind each behavior. Our model achieves this by incorporating intent disentanglement and multi-behavior modeling, facilitated by a parameterized heterogeneous hypergraph architecture. Furthermore, we introduce a novel contrastive learning paradigm that adaptively explores the advantages of self-supervised data augmentation, thereby enhancing the model's resilience against data sparsity and expressiveness with relation heterogeneity. To validate the efficacy of MixRec, we conducted extensive experiments on three public datasets. The results clearly demonstrate its superior performance, significantly outperforming various state-of-the-art baselines. Our model is open-sourced and available at: https://github.com/HKUDS/MixRec.

Figures (5)

• Figure 1: MixRec's framework: (a) Our parameterized heterogeneous hypergraph enables relation-aware intent disentanglement, preserving behavior patterns. (b) Node-level contrastive learning captures local collaborative signals for cross-behavior knowledge transfer. (c) Graph-level contrastive learning performs data augmentation via multi-behavior user discrimination.
• Figure 2: Model Ablation Study.
• Figure 3: Performance w.r.t user interaction frequency.
• Figure 4: Investigation on the impact of hyperparameters $d, E, L$ for the proposed MixRec framework.
• Figure 5: Case study of capturing global user dependencies and multi-behavior relationships in latent embedding space.