An Approach for Detecting Dynamic Communities in Social Networks

TL;DR

This paper introduces new methods for detecting and analyzing overlapping and multi-scale communities in dynamic social networks, addressing challenges posed by temporal changes and overlapping memberships.

Contribution

It proposes OLCPM, an online algorithm for overlapping community detection, and a novel approach for analyzing communities across multiple temporal scales.

Findings

OLCPM effectively detects overlapping communities in dynamic networks.

The multi-scale analysis method identifies stable communities over different time granularities.

Validated on synthetic and real-world datasets, showing practical effectiveness.

Abstract

Recent developments in the internet and technology have made major advancements in tools that facilitate the collection of social data, opening up thus new opportunities for analyzing social networks. Social network analysis studies the patterns of social relations and aims at discovering the hidden features embedded in the structure of social networks. One of the most important features in social networks is community structure : densely knit groups of individuals. The dynamic nature of interaction in social networks often challenges the detection of such community structures. The contributions in this thesis fall into two categories.The first category highlights the problem of identifying overlapping communities over time. To carry out such analysis, a framework called OLCPM (Online Label propagation and Clique Percolation Method) is proposed. It is an online algorithm based on clique percolation and label propagation methods. OLCPM has two main features : the first one is its ability to discover overlapping communities, while the second is its effectiveness in handling fine-grained temporal net works. As for as the second category is concerned, it emphasizes on the problem of analyzing communities that are embedded at different temporal scales. For example, in networks of interaction such as e-mails or phone calls, individuals are involved in daily as well as occasional conversations. We propose a first method for analyzing communities at multiple temporal scales. Hence, the dynamic network (link streams) is studied at different temporal granularities, and coherent communities (called stable communities) over a period of time are detected at each temporal granularity. The two proposed approaches are validated on both synthetic and real-world datasets.