Modelagem de tópicos no estudo da preferência de perfis robôs em tweets relativos a CPI da Covid

Abstract

Twitter is a social network that falls into the microblog category, that is, it allows its users to send and receive short messages through texts and images, within this category concentrates the greater number of users. For this reason, this network has strong relevance in multiple aspects of contemporary society, including in politics. Aiming at this scenario of influence and social impact, many users resort to robot profiles (or bots) to automatically trigger hundreds or even thousands of messages (tweets). This work aims to apply the Latent Dirichlet Allocation (LDA) to identify if there is a preference of robot users for any of the modeled topics. Thus, data collection was carried out through the Twitter API, over approximately 1 month, with the keywords CPI and COVID. 459,145 tweets were collected in Portuguese from 109,027 different users. These users were subject to the analysis of the “Pegabot” platform, which assigns a probability that a given account is or is not a robot. After the pre-stage processing, only users who posted 100 or more messages during the period were selected, remaining 26,966 observations from 189 different accounts. Through the use of LDA it was possible to identify 4 topics, the first topic is associated with the ministry of health, since the second refers to senator Renan Calheiro, the third one focuses on Bolsonaro, while the fourth is related to government in general. In all topics robots had the highest number of publications. It was only found that topic 1 was the one with the lowest proportion of users robots.