Abstract:Recently there is a growing focus on graph data, and multi-view graph clustering has become a popular area of research interest. Most of the existing methods are only applicable to homophilous graphs, yet the extensive real-world graph data can hardly fulfill the homophily assumption, where the connected nodes tend to belong to the same class. Several studies have pointed out that the poor performance on heterophilous graphs is actually due to the fact that conventional graph neural networks (GNNs), which are essentially low-pass filters, discard information other than the low-frequency information on the graph. Nevertheless, on certain graphs, particularly heterophilous ones, neglecting high-frequency information and focusing solely on low-frequency information impedes the learning of node representations. To break this limitation, our motivation is to perform graph filtering that is closely related to the homophily degree of the given graph, with the aim of fully leveraging both low-frequency and high-frequency signals to learn distinguishable node embedding. In this work, we propose Adaptive Hybrid Graph Filter for Multi-View Graph Clustering (AHGFC). Specifically, a graph joint process and graph joint aggregation matrix are first designed by using the intrinsic node features and adjacency relationship, which makes the low and high-frequency signals on the graph more distinguishable. Then we design an adaptive hybrid graph filter that is related to the homophily degree, which learns the node embedding based on the graph joint aggregation matrix. After that, the node embedding of each view is weighted and fused into a consensus embedding for the downstream task. Experimental results show that our proposed model performs well on six datasets containing homophilous and heterophilous graphs.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in multi - view graph clustering tasks, the existing methods have poor performance when dealing with heterogeneous graphs (that is, node connections do not necessarily belong to the same category). Most of the existing methods are designed based on the homogeneous graph assumption. These methods are essentially low - pass filters, mainly focusing on the low - frequency information (similar information) in the graph while ignoring the high - frequency information (dissimilar information). However, in heterogeneous graphs, high - frequency information is very important for learning effective node representations. Therefore, the paper proposes an Adaptive Hybrid Graph Filter for Multi - View Graph Clustering (AHGFC), aiming to improve the clustering performance on heterogeneous graphs by combining low - frequency and high - frequency information. Specifically, the main contributions of the paper include: 1. **Effective utilization of low - frequency and high - frequency information**: An adaptive hybrid graph filter is proposed, which can adaptively mine low - frequency and high - frequency information according to the homophily degree of the graph, thereby preventing information loss and promoting the learning of node representations. 2. **Graph joint aggregation matrix**: A graph joint aggregation matrix is designed, which enhances the distinction between low - frequency and high - frequency signals, making subsequent filtering operations more effective. 3. **Adaptive mechanism**: An adaptive mechanism related to the homophily degree is introduced. The homophily ratio is calculated through pseudo - label information and neighborhood relationship information, thereby dynamically adjusting the weights of low - pass and high - pass filters. 4. **Experimental verification**: Experiments were carried out on multiple datasets, including homogeneous graphs and heterogeneous graphs. The results show that the proposed AHGFC model significantly outperforms the existing methods in performance, especially on heterogeneous graphs. Through these innovations, the paper effectively solves the information loss problem of existing methods when dealing with heterogeneous graphs and improves the performance of multi - view graph clustering.

Homophily-Related: Adaptive Hybrid Graph Filter for Multi-View Graph Clustering

Adaptive Graph Fusion Learning for Multi-View Spectral Clustering

SiMilarity-Enhanced Homophily for Multi-View Heterophilous Graph Clustering

Beyond Homophily and Homogeneity Assumption: Relation-Based Frequency Adaptive Graph Neural Networks

Beyond Homophily: Reconstructing Structure for Graph-agnostic Clustering

Provable Filter for Real-world Graph Clustering

Dual-Optimized Adaptive Graph Reconstruction for Multi-View Graph Clustering

Restructuring Graph for Higher Homophily via Adaptive Spectral Clustering

Dual-Frequency Filtering Self-aware Graph Neural Networks for Homophilic and Heterophilic Graphs

Heterophilic Graph Neural Network Based on Spatial and Frequency Domain Adaptive Embedding Mechanism

Feature Weighted Multi-View Graph Clustering

Finding Global Homophily in Graph Neural Networks When Meeting Heterophily.

Multi-View Bipartite Graph Clustering With Coupled Noisy Feature Filter

Learn from Heterophily: Heterophilous Information-enhanced Graph Neural Network

Revisiting Heterophily For Graph Neural Networks

Homophily-enhanced Structure Learning for Graph Clustering

Addressing Heterogeneity and Heterophily in Graphs: A Heterogeneous Heterophilic Spectral Graph Neural Network

Graph Contrastive Learning under Heterophily via Graph Filters

Graph Filter-based Multi-view Attributed Graph Clustering

PC-Conv: Unifying Homophily and Heterophily with Two-fold Filtering

Exploiting Neighbor Effect: Conv-Agnostic GNN Framework for Graphs with Heterophily