AI for social good

Abstract

This project aims to look at novel ways to detect suicide ideation. Suicide rates have been steadily increasing throughout the years. A wealth of information has also been made available to us with the advent of social media. This project aims to determine the feasibility of using Natural Language Processing, specifically a combination of different pre-trained transformer models and rule-based algorithms, to identify if a particular person displays suicidal ideation. I split the symptoms of suicide into their different categories and experimented with different pre-trained transformer model to find the best model to detect the various symptoms using semantic similarity with cosine similarity. I then proceeded to find the ideal weightage of the categories to maximize the accuracy of my model.