A methodology for community detection in Twitter

Abstract

The microblogging service Twitter is one of the world's most popular online social networks and assembles a huge amount of data produced by interactions between users. A careful analysis of this data allows identifying groups of users who share similar traits, opinions, and preferences. We call community detection the process of user group identification, which grants valuable insights not available upfront. In order to extract useful knowledge from Twitter data many methodologies have been proposed, which define the attributes to be used in community detection problems by manual and empirical criteria - oftentimes guided by the aimed type of community and what the researcher attaches importance to. However, such approach cannot be generalized because it is well known that the task of finding out an appropriate set of attributes leans on context, domain, and data set. Aiming to the advance of community detection domain, reduce computational cost and improve the quality of related researches, this paper proposes a standard methodology for community detection in Twitter using feature selection methods. Results of the present research directly affect the way community detection methodologies have been applied to Twitter and quality of outcomes produced.