On the Homophily of Heterogeneous Graphs: Understanding and Unleashing

2025-01-24

AI Summary

Traditional homophily assumptions fail in heterogeneous graphs (HGs), and cross-type node similarity is difficult to quantify, limiting the performance of heterogeneous graph neural networks (HGNNs). To address this, we propose the **Cross-Type Homophily Ratio (CTHR)**—the first metric to quantitatively characterize label consistency among semantically distinct node types in HGs. Building upon CTHR, we introduce the first **homophily-driven heterogeneous graph structural pruning framework**, which guides redundant edge removal and structural refinement. Evaluated on five real-world HG datasets, our method improves HGNN performance by an average of 13.36%, significantly enhancing generalization. Our core contributions are: (1) formal definition and empirical validation of cross-type homophily; (2) a principled, interpretable, and optimization-friendly homophily metric and pruning paradigm; and (3) a novel perspective and practical toolkit for heterogeneous graph structure learning.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Homophily, the tendency of similar nodes to connect, is a fundamental phenomenon in network science and a critical factor in the performance of graph neural networks (GNNs). While existing studies primarily explore homophily in homogeneous graphs, where nodes share the same type, real-world networks are often more accurately modeled as heterogeneous graphs (HGs) with diverse node types and intricate cross-type interactions. This structural diversity complicates the analysis of homophily, as traditional homophily metrics fail to account for distinct label spaces across node types. To address this limitation, we introduce the Cross-Type Homophily Ratio, a novel metric that quantifies homophily based on the similarity of target information across different node types. Furthermore, we introduce Cross-Type Homophily-guided Heterogeneous Graph Pruning, a method designed to selectively remove low-homophily crosstype edges, thereby enhancing the Cross-Type Homophily Ratio and boosting the performance of heterogeneous graph neural networks (HGNNs). Extensive experiments on five real-world HG datasets validate the effectiveness of our approach, which delivers up to 13.36% average relative performance improvement for HGNNs, offering a fresh perspective on cross-type homophily in heterogeneous graph learning.

Problem

• Heterogeneous Graphs
• Node Similarity
• Graph Neural Networks

Innovation

• Cross-Type Homogeneity Ratio
• Heterogeneous Graph Learning
• Graph Neural Network Enhancement