Automatic Evaluation of Students’ Discussion Skill Based on their Heart Rate

Abstract

Discussion which consists of several Q&A segments (question and answer pairs) is often considered as one of the most familiar types of intellectual and creative activities at meetings. Evaluating students’ answer-quality of Q&A segments in discussion and giving them feedback can effectively help them improve their discussion skills. Considering that the discussion process is a type of cognitive activity, which could result in changes in certain physiological data, such as heart rate (HR) variability (HRV). In this study, we argue that students’ HR data can be used to effectively evaluate the answer-quality of their Q&A segments and as a method of automatic evaluation of students’ discussion-skill compared with using traditional NLP such as semantic analysis. In order to confirm this, we used a non-invasive device, i.e., Apple Watch, to collect real-time updated HR data of students during their discussions in our lab-seminar environment, their HR data were analyzed based on Q&A segments, and three machine-learning models were generated for evaluation: logistic regression, support vector machine, and random forest. The significant HR and HRV features (metrics) were also discussed using a feature selection method. Comparative experiments were conducted involving semantic data of Q&A statements alone and a combination of HR and semantic data. We also gave an experimental investigation on HR and HRV features robustness on the new data set we collected additionally.