[5Yin-A] Poster 8

[5Yin-A-39] Analyzing the Impact of Multimodal Features on Learner Satisfaction in Long-duration Training

〇Kaede Mori¹, Taishi Watanabe², Yusuke Yamamoto², Hayate Funakura^1,3, Ryo Kinoshita¹ (1. Kikagaku, Inc., 2. Ricoh Company, Ltd., 3. Keio University)

Keywords:

AI in Education,Corporate Training,Multimodal Features

Automated evaluation of lecture videos has primarily advanced in the context of short-form asynchronous content, such as MOOCs. However, long-duration synchronous training, such as corporate technical seminars, remains quantitatively under-explored. This study conducts a regression analysis on learner satisfaction using 8-hour training videos provided by Kikagaku, Inc. We integrated linguistic data from transcriptions with visual features from facial expressions and prosodic features from audio waveforms in a multimodal framework. The analysis reveals that even in long-duration training, paralinguistic and non-verbal features—such as vocal inflection and facial dynamics—serve as critical predictors of learner satisfaction, extending beyond the influence of purely semantic information.