EXPLORING OVERLAPPING COMMUNITIES AND THE EFFECT OF OVERLAPPING NODES IN DIFFUSION

Abstract

The evaluation and proliferation in use of online social media platforms allows people to connect and share information. Analyzing these networks enables us to have insight of network spectral and evolutionary dynamics. Analyzing these networks has diverse applications such as link prediction, anomaly detection, marketing influence maximization and recommender systems. The community detection in social network is one such analysis technique that elementally focuses on forming groups or communities based on similar traits or characteristics. Finding communities in social networks has many applications such as finding clusters in network, designing chips and summarizing networks and privacy. Abundant works has been done by many researchers to detect disjoint communities. However, the communities in a given network can be overlapping as well as non-overlapping, such as scientists working on one or more studies can be designated as overlapping entities whereas, the scientist working on their individual studies are designated as non-overlapping communities. In this work, we propose to find the overlapping communities and nodes in social networks based on the property of Louvain [3] and Leiden algorithms [48]. The elementary focus of our proposed community detection algorithm is to generate the confidence matrix, which will portray the confidence of ’belongingness’ [32] in the communities that is computed using the Leiden algorithm [48]. Moreover, with the help of overlapping communities, we detect overlapping nodes. Furthermore, we utilize these detected overlapping nodes for diffusion of infection in given networks. The diverse experimental represent the effect of overlapping nodes in the diffusion process using the cascade process where a node diffuses infection to its immediate neighbours that in turn passes on infection to its immediate neighbours and this process continues until the entire network gets infected. The diffusion dynamics is one such social construct that can be utilized to spread some marketing ideas and practices that can further be utilized to find out how a pathogen can affect a network over a timeline. Moreover, there are also some instances where we can observe that a piece of information gets viral in some time while some information struggles to get noticed. the diffusion dynamics can help us in analyzing these situations. The applications of diffusion dynamics are not only limited to the analysis of spreading/ diffusion but also it can be utilized to detect super spreaders present in a network, that can further be used to model epidemics such as in case of SIS, SI, and SIR models. Finding contact sub-networks and nodes during an epidemic spread can be analysed and detected more efficiently with the help of diffusion dynamics.