An NLP-Based Intelligent Model Approach for Suicidal Tendency Classification

Abstract

Suicidal ideation not only drives a person towards suicide but also helps to commit suicide. Statistics reveal a gradual increase in suicide-related deaths worldwide. This poses a significant threat to both our current and future societies. That is why we need to take necessary measures to prevent it. Nowadays, online media serves us common platform for people to share their feelings and other personal information. In this research, we have gathered data from social media, mainly Reddit, to analyze the suicidal thoughts in the posts. We have investigated some machine and deep learning models for the detection of suicidal ideation in textual data. We have proposed a “DistilBERT-BiLSTM” model by which we can detect textual data for suicidal ideation. The models we looked at and compared were Logistic Regression, Random Forest, Naïve Bayes, Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), Convolutional Neural Network (CNN), CNN-GRU with Attention, Bi-LSTM with Attention, and CNN-GRU with Attention. The dataset we used was made up of posts from people who were having suicidal and non-suicidal thoughts. With an AUC score of 1, our proposed “DistilBERT-BiLSTM” model achieved the highest accuracy of 97%.