A Computationally Efficient Framework for Overlapping Community Detection in Large Bipartite Graphs

2025-12-22

AI Summary

Detecting bi-clique percolation communities (BCPCs) in bipartite graphs suffers from prohibitively high time complexity due to exponential blowup in the size of the maximum bi-clique adjacency graph (MBAG). Method: This paper proposes an efficient BCPC detection paradigm that avoids explicit MBAG construction. It introduces (1) the notion of *partial BCPCs*, enabling vertex-level compression of the MBAG; (2) an implicit adjacency modeling and enumeration mechanism based on shared (α,β)-bi-clique adjacency testing; and (3) integration with optimized maximal bi-clique mining to ensure scalability. Contribution/Results: Experiments demonstrate that the method achieves nearly three orders of magnitude speedup over state-of-the-art BCPC algorithms, while preserving detection accuracy and overlapping community expressiveness. It significantly enhances the practicality and efficiency of overlapping community discovery in large-scale bipartite graphs.

Technology Category

• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Community question answering
• Social Networks and Social Media: Social mining and social search on the Web

Abstract

Community detection, which uncovers closely connected vertex groups in networks, is vital for applications in social networks, recommendation systems, and beyond. Real-world networks often have bipartite structures (vertices in two disjoint sets with inter-set connections), creating unique challenges on specialized community detection methods. Biclique percolation community (BCPC) is widely used to detect cohesive structures in bipartite graphs. A biclique is a complete bipartite subgraph, and a BCPC forms when maximal bicliques connect via adjacency (sharing an (alpha, beta)-biclique). Yet, existing methods for BCPC detection suffer from high time complexity due to the potentially massive maximal biclique adjacency graph (MBAG). To tackle this, we propose a novel partial-BCPC based solution, whose key idea is to use partial-BCPC to reduce the size of the MBAG. A partial-BCPC is a subset of BCPC. Maximal bicliques belonging to the same partial-BCPC must also belong to the same BCPC. Therefore, these maximal bicliques can be grouped as a single vertex in the MBAG, significantly reducing the size of the MBAG. Furthermore, we move beyond the limitations of MBAG and propose a novel BCPC detection approach based on (alpha, beta)-biclique enumeration. This approach detects BCPC by enumerating all (alpha, beta)-bicliques and connecting maximal bicliques sharing the same (alpha, beta)-biclique, which is the condition for maximal bicliques to be adjacent. It also leverages partial-BCPC to significantly prune the enumeration space of (alpha, beta)-biclique. Experiments show that our methods outperform existing methods by nearly three orders of magnitude.

Problem

• Detects overlapping communities in large bipartite graphs efficiently.
• Reduces computational complexity of maximal biclique adjacency graph.
• Prunes enumeration space using partial-BCPC for faster detection.

Innovation

• Uses partial-BCPC to shrink maximal biclique adjacency graph
• Enumerates (alpha, beta)-bicliques to connect maximal bicliques directly
• Leverages partial-BCPC to prune (alpha, beta)-biclique enumeration space