EPPC-OASIS: AI Framework for Mining Patient-Provider Messages
The recently introduced AI framework, EPPC-OASIS, enhances the automated extraction of significant communication behaviors from secure messages between patients and providers. This system integrates ontology-aware adaptation with inference refinement to maintain detailed code and sub-code structures while anchoring annotations in the message text. It incorporates a Wasserstein alignment objective to improve supervised fine-tuning, aligning model representations with the EPPC ontology. Additionally, it employs verification, self-consistency, hybrid correction, and selection or ensembling techniques to tackle remaining prediction inaccuracies. This research is documented on arXiv (2605.24172) and seeks to facilitate the scalable analysis of electronic communication between patients and providers.
Key facts
- EPPC-OASIS is an ontology-aware adaptation approach for structured EPPC extraction.
- It augments supervised fine-tuning with a Wasserstein alignment objective.
- Inference refinement uses verification, self-consistency, hybrid correction, and selection or ensembling.
- The framework addresses challenges in preserving fine-grained code/sub-code structure.
- It grounds annotations in message text.
- The work is published on arXiv with ID 2605.24172.
- EPPC stands for Electronic Patient-Provider Communication.
- The goal is to enable scalable characterization of communication behaviors.
Entities
Institutions
- arXiv