ISCoDe: A framework for interest similarity-based community detection in social networks

This paper proposes a framework for node clustering in computerized social networks according to common interests. Communities in such networks are mainly formed by user selection, which may be based on various factors such as acquaintance, social status, educational background. However, such selection may result in groups that have a low degree of similarity. The proposed framework could improve the effectiveness of these social networks by constructing clusters of nodes with higher interest similarity, and thus maximize the benefit that users extract from their participation. The framework is based on methods for detecting communities over weighted graphs, where graph edge weights are defined based on measures of similarity between nodes' interests in certain thematic areas. The capacity of these measures to enhance the sensitivity and resolution of community detection is evaluated with concrete benchmark scenarios over synthetic networks. We also use the framework to assess the level of common interests among sample users of a popular online social application. Our results confirm that clusters formed by user selection have low degrees of similarity; our framework could, hence, be valuable in forming communities with higher coherence of interests.