Predicting Mental Health of Scholars Using Contextual Word Embedding

Abstract

Frustration, anger and mental fatigue are some of the most prevalent issues faced by scholars these days which are often due to the cut-throat competition and peer pressure among adolescents increasing day by day. The objective of the work was to predict the mental health of scholars by tracking strongly negative words or offensive slangs used very deliberately by them in their tweets. The data to analyze the sentiments exhibited by them has been used from their Twitter timelines. The most common topics they talked about and the common keywords for each of these topics were identified and thereafter using BERT, the word embedding and cosine similarities had been found between these keywords and a bag of words that contained a number of strongly negative emotion words collected manually. In this paper, contextual word embedding has been done on the twitter data of frustrated individuals to analyze their temperament and behavior exhibited by them. Also, we have found if there was any correlation between frequently used negative words and frustrated individuals or not. We then found that there was a correlation and individuals using such words had negative emotions prevalent among them.