One Node One Model: Featuring the Missing-Half for Graph Clustering

TL;DR

This work addresses the under-explored role of node feature information in graph clustering by introducing Feature Personalized Graph Clustering (FPGC) and a per-node 'one node per-collection' approach. Central to FPGC are a squeeze-and-excitation block that selects cluster-relevant features and a feature-cross augmentation within a two-view contrastive framework, producing per-node representations through a lightweight, shared model: $Y = g(X^n) \odot \bar{X}W$ with $\bar{X}=A^kX$. The method is theoretically motivated and empirically validated across diverse datasets, achieving state-of-the-art results and demonstrating robustness to graph quality and scalability to large graphs. Practically, FPGC provides a plug-and-play mechanism to boost various GNN-based clustering pipelines by incorporating feature-perspective personalization and low-order feature interactions.

Abstract

Most existing graph clustering methods primarily focus on exploiting topological structure, often neglecting the ``missing-half" node feature information, especially how these features can enhance clustering performance. This issue is further compounded by the challenges associated with high-dimensional features. Feature selection in graph clustering is particularly difficult because it requires simultaneously discovering clusters and identifying the relevant features for these clusters. To address this gap, we introduce a novel paradigm called ``one node one model", which builds an exclusive model for each node and defines the node label as a combination of predictions for node groups. Specifically, the proposed ``Feature Personalized Graph Clustering (FPGC)" method identifies cluster-relevant features for each node using a squeeze-and-excitation block, integrating these features into each model to form the final representations. Additionally, the concept of feature cross is developed as a data augmentation technique to learn low-order feature interactions. Extensive experimental results demonstrate that FPGC outperforms state-of-the-art clustering methods. Moreover, the plug-and-play nature of our method provides a versatile solution to enhance GNN-based models from a feature perspective.

Figures (7)

• Figure 1: Visualization of results on Cora. (a) is the feature distribution of different clusters, which shows that clusters are characterized by different features. (b) is the DTW distance matrix based on cluster-relevant features, which verify their distinctiveness. We can draw the conclusion that the cluster-relevant features contain valuable information about clusters.
• Figure 2: The pipeline of our FPGC. We preprocess the node features by stacked graph filters. Besides, we also input original features into the squeeze-and-excitation block to select the top $n$ significant features, based on which we learn a model for each node. Then, the contrastive framework encodes the smoothed node features and augmented features to achieve discriminative node representations.
• Figure 3: Parameter analysis of $m$ and $n$ on Cora and EAT.
• Figure 4: Parameter analysis of $k$ and $\lambda$ on Cora and EAT.
• Figure 5: Clustering results and running time on Flickr and Twitch-Gamers.

Theorems & Definitions (1)

• Theorem 1