Tag
This paper introduces interpretable form-level features to detect and guide LLM-generated Korean poetry, achieving improved detection accuracy and generating poems that more closely resemble human writing in form.
This paper introduces an LM-guided counterfactual recommendation pipeline for improving doctor-patient communication in text-based telemedicine. It identifies interpretable features like tone and actionability, and suggests minimal changes that increase positive patient feedback without altering medical content, achieving a mean 6.41% gain in predicted positive feedback.