Detectando comportamento automatizado nos tópicos de tendência do Twitter no Brasil

Abstract

The growth in the number of users in social networks, especially Twitter, become themselves susceptible to creation and propagation of automated posts. On Twitter, the Trend Topics list represents the most talked subjects in a particular region and can be misused by automated accounts. Then, it is necessary to understand and study how these users behave in order to create measures to combat them and ensure that published data have credibility. Using a real database collected from the Twitter Trend Topics in Brazil, from December 2013 to June 2014, with 2.853,822 accounts and 11,294,861 tweets, a methodology to detect automated behavior in Trend Topics was proposed. For this, we studied several text characteristics and user behavior to identify attributes capable of distiguish human users and automated users. Also were proposed six (6) new features based on the concept of entropy. Using this set of attributes with ma-chine learning algorithms for supervised classification, it was possible to detect 92 % of automated accounts in the database used and thus get an insight into the behavior of these users.