Abstract
Parkinson's disease (PD) drives patients and care partners to online communities, but forums often misalign help-seekers' needs with replies (e.g., empathy requests met with links). We frame this as a socio-technical matching problem and adopt a design science research approach rooted in the Social Support Behavior Code and speech act theory. We contribute four artifacts: (1) a transformer-based, multi-label classifier that infers five support intents (informational, emotional, instrumental, esteem, appraisal); (2) a response protocol mapping detected intents to concrete communicative moves; (3) a prototype interface that surfaces visual intent cues and provides inline scaffolds (prompts, checklists, phrasings); and (4) prescriptive design principles for social-support systems. Evaluation ranges from formative model metrics (multi-label F-scores, calibration) to ecosystem outcomes (time-to-aligned-response, fewer "link-dump" replies, perceived helpfulness, newcomer integration), with clear safeguards distinguishing peer support from medical advice. This position supports the alignment of intent as a tractable, theory-informed target for AI mediation in PD and related conditions.
Recommended Citation
Venkatesan, Srikanth; Mulgund, Pavankumar; and Deogaonkar, Anagha Shrikant, "AI-Enabled Intent Detection and Response Scaffolding in Parkinson’s Communities: A Design Science Study" (2025). Proceedings of the 2025 Pre-ICIS SIGDSA Symposium. 75.
https://aisel.aisnet.org/sigdsa2025/75