A Deep Learning Suicide Ideation Using BERT Model

Abstract

Suicidal Ideation in society today has become very common, this is due to stressful societal issues. Recently social platforms have gained special attention regarding this phenomenon. Mental health issues like depression, frustration, hopelessness, and bullying among others directly or indirectly influence suicidal thoughts. Early detection of suicidal intent can help people to diagnose and get proper treatment before it is too late. In this study, a novel detection approach that uses a deep learning approach is proposed. Essentially the study analyses raw language data from different sources such as social networks among others, and classifying the indication of suicidal ideation. This study focuses on Deep learning techniques as a base for suicidal ideation. To conduct this experiment, we employed a locally gathered dataset in Kenya, which was labeled to distinguish between instances with suicidal indications and those without. The experiment shows that the model can achieve an optimal classification result. The BERT model surpassed the performance of traditional Naïve Bayes models in measures of precision, recall, and F1 score.