An analysis of community structure in Brazilian political topic-based Twitter networks

Online social networks such as Twitter are important platforms for spreading public opinion on a variety of subjects. The classification of users through the analysis of their posts on Twitter according to their opinion sharing can help marketing ads and political campaigns to focus on specific user groups. Community detection-based techniques are specially useful to classify Twitter users, as they do not require rule-based methods or labeled data to perform the clustering task. In this paper, we constructed networks using data related to political discussions in Brazil extracted from Twitter. We show that (i) these networks follow the power-law distribution, indicating that a few popular users are responsible for most of the "mentions" and "retweets"; (ii) the most popular tweets are viral and spread across the communities whereas most of the remaining tweets are trapped in the communities where they originated; and (iii) words associated with positive sentiments are predominant in network communities related to the Brazilian presidential elections and appear in viral tweets.