Detecting semantic overlapping communities and influential nodes in social networks

Abstract

Social network communities are composed of people with common interests who influence or are influenced by themselves. Identifying and exploring these relationships are fundamental activities to support collaboration in the network. In the present work we analyze the level of influence among people and their communities, by analyzing social networks considering both their structure and context information. To this, complex networks concepts and semantic technologies are used to combine linkage-based analysis with content-based analysis in social networks. We first propose an algorithm for detecting overlapping communities and important nodes in the network, named NetSCAN. Then, an ontology called NetO is proposed aiming to support semantic analysis. The proposed algorithm and ontology were first tested separately in controlled experiments based on well known datasets. Then, in two history research evaluations, they were used together to perform structural and semantic analysis over two real-world social networks. In first evaluation a Scientific Social Network was analyzed, based on a large scientific repository called DBLP. The second evaluation analyzed a Software Development Social Network constructed from StackOverow data, which is one of the most popular question-answer (Q&A) forum in this context. The first evaluation has shown that we were able to detect scientific communities, influential researchers and their research interests. In the second evaluation, communities of software developers were detected as well as expert developers and their topics of expertise. Therefore, the results points to the viability and effectiveness of the proposed solution.