Word Embeddings and Machine Learning Classifiers Applications for Automatic Detection of Suicide Tendencies in Social Media

Abstract

This study presents an innovative and comprehensive model for the automatic detection of suicidal ideation in social media posts. Through an in-depth analysis of 50000 posts and the combination of four word embedding techniques (Word2Vec, GloVe, MPNet, and GPT-3) with five advanced classifiers, we have achieved an accuracy of over 90% in identifying users who may be at risk. Our results suggest that the integration of large language models like GPT-3 with deep neural network architectures offers a promising tool for suicide prevention in the digital realm, contributing to the development of automated screening systems capable of alerting mental health professionals to potential cases of risk.