Towards A Machine Learning Framework for Suicide Ideation Detection in Twitter

A worldwide rise of number of individuals had been observed to commit suicide at young age since 2019. With emergence of digital technologies, more suicides attempts were seen to have been pre-communicated through social media posts before the attempts. As such, much initiatives had been conducted to detect communication of distress on social media which imply risks of suicide attempts or any related forms of self-harm. Such initiatives aim to predict the possibility of suicide attempts for prevention measures. This work proposes a framework for machine learning of suicide ideation detection in social media which will contribute to the prediction of possibility of suicide attempt in a social media post. The framework includes data acquisition, data annotation, data pre-processing, feature extraction, classification and performance evaluation of the machine learning model. Acquisition of social media post samples from Twitter is proposed for training and testing datasets through representational state transfer application programming interface. Manual annotation is conducted on the training datasets with different formulated suicide risks. Data pre-processing is done through natural language toolkit and performance evaluation through valence aware dictionary for sentiment reasoning. This work will be further extended with a web-based application for implementation and evaluation of the trained machine learning model.