H-Louvain: Hierarchical Louvain-based community detection in social media data streams

Abstract

The rise of online social networks has fundamentally transformed the traditional way of social interaction and information dissemination, leading to a growing interest in precise community detection and in-depth network structure analysis. However, the complexity of network structures and potential issues like singularity and subjectivity in information extraction affect the accuracy of community detection. To overcome these challenges, we propose a new community detection algorithm, known as the Hierarchical Louvain (H-Louvain) algorithm. It enhances the performance of community detection through a multi-level processing and information fusion strategy. Specifically, the algorithm integrates graph compression techniques with the Hyperlink-Induced Topic Search (HITS) algorithm for initial network hierarchical partitioning, simultaneously filtering out low-quality posts and users while retaining critical information. Furthermore, the proposed method enhances semantic representation by automatically determining an appropriate number of attribute vector dimensions and obtaining attribute weight information through the calculation of self-authority values and the "minimum distance" attribute of posts. Lastly, the method creates an initial user training set through network re-partitioning in hierarchical layers and improves the Louvain algorithm for community partitioning by estimating the comprehensive influence of nodes. Extensive experimentation has demonstrated that the H-Louvain algorithm outperforms state-of-the-art comparative algorithms in terms of accuracy and stability in community detection based on real-world Twitter datasets.