Keyword-Driven Depressive Tendency Model for Social Media Posts

Abstract

People are increasingly sharing posts on social media (e.g., Facebook, Twitter, Instagram) that include references to their moods/feelings pertaining to their daily lives. In this study, we used sentiment analysis to explore social media messages for hidden indicators of depression. In cooperation with domain experts, we defined a tendency towards depression as evidenced in social media messages based on DSM-5, a standard classification of mental disorders widely used in the U.S. We also developed three data engineering procedures for the extraction of keywords from posts presenting a depressive tendency. Finally, we created a keyword-driven depressive tendency model by which to detect indications of depression in posts on a major social media platform in Taiwan (PTT). The performance of the proposed model was evaluated using three keyword extraction procedures. The DSM-5-based procedure with manual filtering resulted in the highest accuracy (0.74).