[1Yin-A] Poster 1

[1Yin-A-40] A Platform-Based Framework for Post-Hoc Positive Feedback Generation in Collaborative Interaction Analysis

〇Masaki SHUZO¹, Kazuaki Kondo², Kei Shimonishi², Atsuo Tsuchiya¹, Yoshimi Miyata¹, Kenta Abe³, Motoki Sakai⁴ (1. Shonan Institute of Technology, 2. Kyoto University, 3. Teikyo University, 4. Nihon University)

Keywords:

Collaborative interaction,Positive feedback,Large Language Models

This study aims to establish a foundation for collaboration interaction analysis and for generating post-hoc positive feedback (PFB) by constructing a multimodal dataset based on the “Paper Tower” collaborative task. We synchronously recorded video, audio, IMU, biological signals, and questionnaires to support multifaceted analyses of interaction phenomena such as non-task-oriented utterances, overlapping speech, and task engagement. Using exploratory analyses, we observed tendencies suggesting social roles of non-task-oriented utterances in team building, and we examined how the frequency of overlapping speech relates to the stability of task performance. We also illustrate how an “Engagement Map,” which visualizes time-series engagement states, can be used as evidence when drafting PFB messages. These observations provide practical guidelines for connecting multimodal sensing to evidence-grounded reflective support. Future work includes organizing an analysis challenge based on shared evaluation metrics using this dataset.