Sentiment Analysis on Social Media Commentaries To Forestall Suicidal Ideation

Philip Heo
‘Iolani School

Abstract

Sentiment analysis is defined as “a natural language processing technique used to determine whether data is positive, negative or neutral.”1 It is often used as a method to analyze the mental or emotional state of a person who might be going through mental struggles and facing personal challenges that urge them to make unhealthy or suicidal decisions. In order to prevent melancholic disorders or others of the sort, we used machine learning techniques to predict and analyze the emotions of the person by looking at that person’s comments posted on the internet, specifically on social media platforms.
We modelled a general twitter comment, processing it, and applied it with a naive bayesian model and confirmed the operation of the program. However, there were limits to the naive bayesian model of not being able to consider the word order of the sentences inputted. From this, a long short-term memory(LSTM) and recurrent neural network (RNN) were used to take into account the words in the sentences in a more precise fashion. Since the sentences contain many different forms of the same word, prepositions, and slangs, a preprocessing of removing unnecessary words and creating a list of commonly used slangs was implemented. This ultimately yielded an 81% of accuracy overall.

Please click to read the paper.