WOCD: A Semi-Supervised Method for Overlapping Community Detection Using Weak Cliques Journal: Knowledge-Based Systems

Shaozhen Ma, Hanchen Wang, Dong Wen, Wenjie Zhang, Wei Huang, Ying Zhang · 2025

Read WOCD: A Semi-Supervised Method for Overlapping Community Detection Using Weak Cliques Journal: Knowledge-Based Systems by Shaozhen Ma, Hanchen Wang, Dong Wen, Wenjie Zhang, Wei Huang, Ying Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-00927.txt.

Introduction

Shaozhen Ma and co-authors find overlapping communities in networks by planting seed labels in tight little groups, in a 2025 paper. Many real communities overlap: a person belongs to family, work, and hobby circles at once. Older detectors use links alone; newer neural ones struggle to fuse links, attributes, and prior knowledge, and blur communities as they look wider. Their method, WOCD, works semi-supervised, inventing training labels on the fly.

The seed labels come from weak cliques, groups where members are densely but not perfectly connected, which exploits link structure cheaply. Prior information is folded in from the start. The authors report better accuracy on complex overlaps than representative rivals. Community detection feeds everything from fraud rings to protein modules, so a sturdier detector travels far.

Catalog record

CREATORShaozhen Ma, Hanchen Wang, Dong Wen, Wenjie Zhang, Wei Huang, Ying Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-00927

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

WOCD: A Semi-Supervised Method for Overlapping Community Detection Using Weak Cliques Journal: Knowledge-Based Systems Shaozhen Ma, Hanchen Wang, Dong Wen, Wenjie Zhang, Wei Huang, Ying Zhang Abstract Overlapping community detection (OCD) is a fundamental graph data analysis task for extracting graph patterns. Traditional OCD methods can be broadly divided into node clustering and link clustering approaches, both of which rely solely on link information to identify overlapping communities. In recent years, deep learning-based methods have made significant advancements for this task. However, existing GNN-based approaches often face difficulties in effectively integrating link, attribute, and prior information, along with challenges like limited receptive fields and over-smoothing, which hinder their performance on complex overlapping community detection. In this paper, we propose a Weak-clique based Overlapping Community Detection method, namely WOCD, which incorporates prior information and optimizes the use of link information to improve detection accuracy. Specifically, we introduce pseudo-labels within a semi-supervised framework to strengthen the generalization ability, making WOCD more versatile. Furthermore, we initialize pseudo-labels using weak cliques to fully leverage link and prior information, leading to better detection accuracy. Additionally, we employ a single-layer Graph Transformer combined with GNN, which achieves significant performance improvements while maintaining efficiency. We evaluate WOCD on eight real-world attributed datasets, and the results demonstrate that it outperforms the state-of-the-art semi-supervised OCD method by a significant margin in terms of accuracy. 1 Introduction Community structure is a fundamental characteristic of complex networks, widely present in various fields such as social, biological, and information networks [ 1 , 2 , 3 ] . The community is the group of nodes that are more densely connected compared with the nodes outside the group [ 4 ] . Community detection (CD) aims to uncover the underlying community structure of networks, revealing how nodes form tightly connected groups. Numerous CD methods [ 5 , 6 ] have been successfully applied to identify distinct, non-overlapping community structures, making significant contributions to the understanding of community structures. However, real-world networks often contain overlapping community structures [ 7 ] , where nodes belong to multiple groups. As a result, these overlapping communities cannot be effectively captured by non-overlapping CD methods. In the past decade, researchers have increasingly focused on overlapping community detection (OCD), leading to significant advancements in this field. Algorithm-based methods like CPM [ 8 ] , COPRA [ 9 ] and NMF [ 10 ] are popular solutions to OCD problem. CPM detects communities by merging adjacent k-cliques that share common nodes, while COPRA assigns labels based on neighborhood consensus, adjusting the belonging coefficient to control the degree of community overlap. Meanwhile, NMF-based methods [ 11 , 12 ] derive soft community memberships by factorizing the adjacency matrix into low-rank factor matrices. However, these approaches rely solely on link information, overlooking valuable attributes and prior information, which often results in limited effectiveness and practicality for large-scale networks. NOCD [ 13 ] is the first method that combines graph neural networks (GNNs) with the Bernoulli-Poisson probabilistic model for the OCD problem, achieving improved detection performance. However, employing a two-layer graph convolutional network (GCN) inherently limits its capacity to capture long-range node relationships. Based on NOCD, DynaResGCN [ 14 ] addresses the limitations of shallow GCNs by introducing up to forty layers of deep residual GCN with dynamic dilated aggregation for the OCD problem, but this approach significantly increases computational overhead and the risk of over-smoothing, especially in large-scale networks. Furthermore, most existing OCD methods rely solely on link and attribute information, placing relatively little emphasis on prior information (e.g., known node-community labels). SSGCAE [ 15 ] addresses this limitation by incorporating prior information into its learning process. It employs a convolutional autoencoder that integrates link and attribute information while introducing a semi-supervised loss to enhance detection accuracy. However, it only utilizes prior information as an auxiliary loss term, limiting its ability to fully leverage this valuable external knowledge. Driven by the limitations of existing OCD methods, we identify five key motivations in this work. Fully Utilizing Prior Information. Various CD methods [ 16 , 17 ] have shown that incorporating prior information is critical for detection accuracy. In contrast, most existing OCD methods overlook the utilization of prior information, thereby limiting their overall effectiveness. Making Full Use of Link Information. Algorithmic methods, such as CPM, effectively predict community affiliations for certain nodes by leveraging link information. However, many existing learning-based methods rely solely on neighborhood aggregation within GNNs, leading to incomplete utilization of link information. Performance Instability. In real-world applications, prior information is typically available for only a small subset of nodes, making semi-supervised models a more practical choice. Nevertheless, the incompleteness of prior information negatively affects the generalization ability of these models, leading to performance instability. Limitations of GNNs. Existing GNN-based OCD models are constrained by limited receptive fields [ 18 , 19 ] and over-smoothing [ 20 , 21 ] , lacking efficient solutions to address these issues. For instance, DynaResGCN [ 14 ] attempts to expand the receptive field by stacking up to forty layers of GNN. However, this strategy significantly increases computational overhead and the risk of over-smoothing, especially in large-scale networks. Lack of Semi-supervised Baselines. While previous studies [ 6 , 22 , 15 ] have demonstrated that incorporating limited supervision can enhance model performance, most existing OCD methods remain unsupervised. To the best of our knowledge, SSGCAE is the only available learning-based semi-supervised OCD baseline. This shortage limits comprehensive benchmarking and further development in in the OCD field. Based on the above motivations, we propose a Weak-clique-based Overlapping Community Detection method, namely WOCD. WOCD first assigns high-quality pseudo-labels to a subset of nodes by identifying weak cliques, a form of cohesive subgraph. After that, WOCD employs a novel semi-supervised framework with pseudo-labels to efficiently utilize prior information. Additionally, WOCD uses a single-layer Graph Transformer combined with GNN to improve performance while maintaining efficiency. The main contributions of our work are summarized as follows: 1. We propose a novel semi-supervised framework that integrates link, attribute, and prior information for overlapping community detection. By incorporating pseudo-labels, the framework adapts effectively to conditions with limited labeled data, making it well-suited for real-world applications. 2. We propose a pseudo-label initialization module based on weak cliques. WOCD can generate high-quality pseudo-labels before the training phase by leveraging multi-faceted information within the graph, leading to enhanced generalization ability. 3. We combine a lightweight Graph Transformer with GNN to expand the receptive field and alleviate over-smoothing issue. This combination significantly improves detection accuracy while maintaining efficiency. 4. We address the lack of semi-supervised baselines by effectively incorporating semi-supervised losses into multiple unsupervised OCD methods, which enables a fair comparison with WOCD. 5. We conduct extensive evaluations on eight datasets. Compared with six existing OCD baselines, WOCD shows a significant accuracy advantage. Additionally, ablation studies confirm the performance improvement brought by each component of WOCD. 2 Related Work 2.1 Overlapping Community Detection Traditional overlapping community detection (OCD) methods are typically divided into node clustering and link clustering approaches [ 23 ] . Node clustering detects communities by grouping nodes based on their structural features, focusing on the properties and relationships of individual nodes [ 24 , 25 ] . In contrast, link clustering first

Read the complete text