In plain words
Borderline personality disorder shapes how a person thinks and feels about themselves and others, yet it is notoriously under-diagnosed. This work — done with a practising psychiatrist as co-author — asks whether the disorder leaves a detectable fingerprint in the emotions people express in text.
Two kinds of data were studied: posts from people on social media who self-identify as BPD patients, and responses from people invited to take a BPD screening test. From every text, the emotions expressed were extracted as features and fed to a Random Forest classifier.
The model reached around 81% accuracy — and strikingly, using only the three most diagnostic emotions (nervousness, shame and pain) performed just as well, suggesting BPD carries a distinctive emotional signature.
How it works
- Gather two data sourcesSocial-media posts from self-declared BPD patients, plus screening-test responses from volunteers.
- Extract emotional featuresEach text is converted into the emotions it expresses, rather than raw words.
- Train and rankA Random Forest learns to separate BPD from non-BPD text and ranks which emotions matter most.
- Distil the signatureNervousness, shame and pain alone reproduce ~81% accuracy — a compact emotional marker of the disorder.
Why it matters
A screening aid that works on ordinary text could reach people who would never book a psychiatric evaluation, and give clinicians an early, evidence-backed reason to look closer. It is an assistive signal, not a diagnosis — but for an under-diagnosed disorder, earlier signals save years.