Linguistic trajectories of bipolar disorder on social media
This work addresses scalable monitoring of mental health for clinicians and researchers, but it is incremental as it builds on existing social media analysis methods.
The study tackled the problem of identifying language markers for bipolar disorder (BD) by analyzing social media posts from users before and after diagnosis, showing pervasive linguistic alterations reflecting mood disturbance and other psychiatric factors, with recurring mood-related changes exhibiting a 12-month periodicity.
Language provides valuable markers of affective disorders such as bipolar disorder (BD), yet clinical assessments remain limited in scale. In response, analyses of social media (SM) language have gained prominence due to their high temporal resolution and longitudinal scope. Here, we introduce a method to determine the timing of users' diagnoses and apply it to study language trajectories from 3 years before to 21 years after BD diagnosis - contrasted with uses reporting unipolar depression (UD) and non-affected users (HC). We show that BD diagnosis is accompanied by pervasive linguistic alterations reflecting mood disturbance, psychiatric comorbidity, substance abuse, hospitalization, medical comorbidities, unusual thought content, and disorganized thought. We further observe recurring mood-related language changes across two decades after the diagnosis, with a pronounced 12-month periodicity suggestive of seasonal mood episodes. Finally, trend-level evidence suggests an increased periodicity in users estimated to be female. In sum, our findings provide evidence for language alterations in the acute and chronic phase of BD. This validates and extends recent efforts leveraging SM for scalable monitoring of mental health.