Paper Title
EARLY DETECTION OF MENTAL HEALTH CRISES THROUGH SOCIAL MEDIA ANALYSIS: USING NATURAL LANGUAGE PROCESSING TO IDENTIFY SUBTLE LINGUISTIC CHANGES THAT MAY INDICATE DETERIORATING MENTAL HEALTH

Abstract
Mental health crises represent serious challenges for public health systems everywhere. Early detection and intervention is crucial to mitigate the impact of these crises on individuals and communities. Despite the number of studies performed using NLP techniques to assess social media content over deteriorating mental health, we shed light on ways how these can be used to underlie early identification. There, we collected social media posts from 10,000 users and analyzed the linguistic markers associated with changes in mental health status over a 12-month period. Using a BERT-LSTM model, our NLP model was able to identify users with deteriorating mental health (≡83%) The most significant linguistic cues were greater rates of first-person singular pronouns, negative emotion words and different levels of sentence complexity. These results illustrate the potential significance of NLP for social media analysis in early warning systems, towards impacting earlier intervention and improved patient centered outcomes during mental health crises.