Recent, dramatic increases in life-threatening self-harm and hospitalizations of young people due to mental disorders has motivated efforts to preemptively identify suffering individuals and connect them with intervention. One such approach has been to predict the development of these severe mental disorders in individuals based on their social media activity using Natural Language Processing artificial intelligence models. However, these models haven’t been developed for short-form text social media platforms commonly used by young people. This research project seeks to understand whether Natural Language Processing artificial intelligence models are capable of predicting mental disorders in young users on these different platforms by exploring: 1) topical differences, and 2) linguistic differences across platforms. Natural Language Processing analysis techniques were used to extract meaning from the textual activity of users on short-form social media platforms and predict their mental health. This study found that similar topics were discussed on both types of social media platforms. Although, the linguistic characteristics collected by the Natural Language Processing algorithm were less robust on short-form platforms, this study developed a novel method of combining multiple posts from a single user to achieve similarly high predictive accuracy on short-form platforms. This was achieved by analyzing the changes in social media textual behavior across multiple posts by a single user over time. These findings demonstrate the promising applicability of Natural Language Processing-based predictive models to identify users with mental disorders on short-form social media platforms more common among young people.

Yubi Mamiya, UG '26:  www.linkedin.com/in/yubimamiya