Classificação e clusterização de dados de mídias sociais obtidos através de wb crawlers para análise de sentimentos e ideações suicidas

Abstract

With the rise of social media the advent of the internet, users of these platforms generate a large amount of data that makes it possible to explore them in different formats. Numerous means and techniques help extract information in the web environment and rules for handling and applying machine learning models to analyze and discover patterns in the records. Research has shown that a given user’s posts on social media can express different emotions, or even the message can contain suicidal ideation, whether they are clear or not. Recent works have used sentiment analysis as strategies to identify the polarity of a text. Other studies demonstrate the use of the suicidal risk assessment scale as a tool to define the degree of suicide contained in a given content. This work aims to use automation algorithms to build a database containing as source the social media Facebook, Instagram, Twitter and Mundo Psicólogos. In addition, it focuses on the application of machine learning algorithms to analyze feelings and suicidal ideation. As a result of this study, a base containing 32,794 messages was built through the implementation of web crawlers. Also, 1,600 messages were manually labelled in 5 different groups representing suicidal levels, and 31,974 automatically in positive or negative polarity. This paper used four machine learning algorithms: Random Forest with Naive Bayes for classification on the analysis of feelings and K-means and DBSCAN for clustering on suicidal ideation. Of the algorithms used for analyzing feelings, RandomForest showed the best results according to the metrics used. About clustering, none of the previously defined classes was fully grouped with any of the generated groups.