What type of study is this?

This is a Quantitative Study study (also classified as: Experimental Study).

October 8, 2025Open Access

Community Detection Based on the LO-WT Algorithm

Key Points

The LO-WT algorithm improves community detection by integrating the Louvain algorithm and Walk-Trap algorithm, significantly outperforming traditional methods.
Critical adjustments, such as similarity thresholds and trajectory lengths, were optimized to enhance performance across four different network types.
Results indicated LO-WT surpassed the Louvain algorithm in Normalized Mutual Information, achieving scores up to 0.83 for Facebook network detection.
The methodology shows LO-WT's capabilities in detecting meaningful community structures, affirming its efficiency in network analysis.

Abstract

Community detection is an important and difficult task in network analysis. This study proposes the Louvain algorithm combined with the Walk-Trap algorithm (LO-WT), leveraging structural and path-based similarities to enhance the detection accuracy. The structural similarity between nodes is a sensitive issue and essential for community identification. First, computing the similarity between the nodes is applied to determine the weights assigned to the edges connecting them. On the other hand, this information is used to guide the walking process in the Walk-Trap algorithm to obtain the trajectories for each node. Then, the similarity is calculated based on the Jaccard similarity among the trajectories. This builds a similarity matrix, which is then used to create a new graph as a weighted network based on a certain threshold depending on the network's complexity. The weights represent the degree of similarity between the nodes. The Louvain algorithm can then be effectively applied to the new graph to identify the communities by maximizing the modularity rather than using agglomerative clustering in the Walk-Trap algorithm. This approach leverages the efficiency of the algorithm to uncover the meaningful community structures within the data. Four real and synthetic networks are used to validate the results, and the algorithm is evaluated against several algorithms, including baseline and state-of-the-art methods. Critical parameters, such as the structure-based similarity threshold (0.1–0.3) and trajectory length (2–4) are carefully adjusted to optimize the performance. The results show that LO-WT significantly outperforms recent related methods and traditional algorithms. Specifically, it outperforms the Louvain algorithm in terms of Normalized Mutual Information (NMI) and is competitive regarding modularity. Additionally, it surpasses related work by achieving higher performance across all four real-world networks: Karate (0.41), Dolphin (0.52), Football (0.60), and Facebook (0.83). Furthermore, LO-WT can exhibit high conductivity and density, demonstrating its robust performance. Overall, the LO-WT algorithm demonstrates its effectiveness for accurate community detection.

Community Detection Based on the LO-WT Algorithm

Key Points

Abstract

Cite This Study