Scale-to-Dialogue: Low-Burden Elicitation of Daily Premenstrual Symptom Ratings with Small Language Models

A recent preprint on arXiv (2608.08746) presents Scale-to-Dialogue, an innovative approach utilizing conversational AI to gather daily premenstrual symptom assessments, aiming to lessen the response burden. This method frames the task as an ordinal label-recovery challenge, prompting users to provide feedback on a limited number of symptom clusters, which are then correlated with their original severity ratings. The study analyzed 3,320 participant-days from the mcPHASES dataset, focusing on cramps, mood swings, fatigue, sleep disturbances, stress, and bloating, using a six-point scale. Out of this, six participants were designated for development and 36 for evaluation, resulting in 360 participant-days and 2,160 item labels. The fixed six-item questioning yielded a quadratic weighted kappa of 0.976, while three joint symptom-cluster inquiries achieved 0.913, with 97.45% agreement within one severity level. This research highlights the capabilities of small language models in health evaluations, providing a less burdensome option compared to conventional methods.