From Tweets to Insights: A Deep Learning Framework for Suicide Prediction

Mental health crisis now including suicide has become a global concern with so many difficulties and catastrophes experienced in our societies. Twitter in particular serves as platforms where people share their emotions, let out frustrations and struggles. The analysis of these emotions offers insights on the trend of public mental health thereby allowing the forecasting of emotional predispositions across geographic regions. This paper employs the sentiment analysis technique on deep learning techniques such as LSTM to analyse public sentiments from the data got from Twitter. Providing the information about the reactions towards significant events of individual’s life, for example, natural disasters, or economic crisis, the model might define certain affective profiles which, in turn, can be associated with a higher risk of a person’s suicide. Knowledge of these changes in moods therefore can help officials curb the effect of disasters on the health of a persons’ mind and subsequently decrease the number of suicides. This research proposes a BERT-CNN-LSTM architecture to improve the suicidality prediction performance from SM data. The proposed model which is depicted above has a uncanny ability to analyze complex emotional patterns and provide real time feedback for psychological actions. This approach appears to be more effective in detecting emotional distress as compared to the traditional model for timely preventive actions.