A Noise-Resilient Semi-Supervised Graph Autoencoder for Overlapping Semantic Community Detection

TL;DR

This work tackles overlapping community detection in noisy attributed networks by introducing a semi-supervised graph autoencoder that fuses graph topology, node attributes, and prior knowledge through a graph multi-head attention encoder and a modularity-based objective. The model learns semantic representations, reconstructs topology, and optimizes community structure with a loss that combines reconstruction, semi-supervised guidance, and modularity maximization. It demonstrates superior performance over state-of-the-art methods on real datasets and shows strong robustness to substantial attribute noise, highlighting the value of integrating semantic information with structural patterns. The approach is poised to improve accurate community discovery in complex networks where features are noisy or incomplete, with implications for understanding roles and interactions in social, collaboration, and other networks.

Abstract

Community detection in networks with overlapping structures remains a significant challenge, particularly in noisy real-world environments where integrating topology, node attributes, and prior information is critical. To address this, we propose a semi-supervised graph autoencoder that combines graph multi-head attention and modularity maximization to robustly detect overlapping communities. The model learns semantic representations by fusing structural, attribute, and prior knowledge while explicitly addressing noise in node features. Key innovations include a noise-resistant architecture and a semantic semi-supervised design optimized for community quality through modularity constraints. Experiments demonstrate superior performance the model outperforms state-of-the-art methods in overlapping community detection (improvements in NMI and F1-score) and exhibits exceptional robustness to attribute noise, maintaining stable performance under 60\% feature corruption. These results highlight the importance of integrating attribute semantics and structural patterns for accurate community discovery in complex networks.

Figures (3)

• Figure 1: Architecture of the proposed model for overlapping community detection. Our model integrates attribute and prior information using a graph multi-attention autoencoder based on modularity maximization.
• Figure 2: Sensitivity analysis of model performance on the Facebook, Engineering, and Computer Science datasets under varying noise levels ($P_{\mathrm{mis}}$), with prior information fixed at 10%. Robustness was evaluated using ONMI and F1‑score to assess stability as $P_{\mathrm{mis}}$ increases.
• Figure 3: Performance comparison of all studied models in terms of ONMI and F1-measure on the FB dataset, Engineering dataset, and Computer Science dataset, with $P_{\text{mis}}$ fixed at 60% noise attribution and 02% of prior information.