Improving community detection algorithms in directed graphs with fuzzy measures. An application to mobility networks

Abstract

This paper proposes a novel methodology to enhance any community detection algorithm for directed networks by introducing a flow-based fuzzy measure, which improves both partition quality and the interpretability of the algorithm. To do so, we focus on a novel aggregation paradigm which combines social networks with fuzzy measures. We explore the potential of incorporating information from fuzzy measures, specifically with a flow capacity measure, to improve and optimize community detection algorithms. We present a detailed evaluation process to demonstrate the effectiveness of this methodology. To achieve this, we analyze a robust repository of databases, using several classical community detection techniques. A comprehensive comparison of classic results with the new methodology demonstrates the effectiveness of the presented aggregation paradigm. We reach a more conclusive understanding of the impact of our methodology through the application of machine learning techniques. Therefore, the proposed methodology enhances community detection performance in directed networks by incorporating flow-based fuzzy measures. We also illustrate its effectiveness through a case study. It showcases improved partition quality compared to traditional algorithms, along with theoretical insights into the fuzzy approach.