Ideiagintza suizidaren identifikazioa sare sozialetan

Abstract

Suicide has become one of society’s main concerns in recent years. In addition, social media has become part of our everyday life and is often used to express emotions. In this work, a binary classification has been carried out to determine whether or not the content of a message on the Reddit social network is related to suicide. On the one hand, with regard to supervised systems in the state of the art, the best performance has been achieved with the ELECTRA transformer, with an accuracy-rate of 97.9%. On the other hand, it has been concluded that the representations produced by the LDA topic-model can be useful for this task, and to prove this, a baseline classifier has been proposed, which has reached an accuracy of 83.3%.