Evaluation of Automatic Collaborative Learning Process Coding Using Deep Learning Methods Based on Multi-Dimensional Coding Scheme

Chihiro Shibata, Kimihiko Ando, Taketoshi Inaba

Abstract: In computer-supported collaborative learning research, it is important to determine the guidelines for appropriate scaffolding by extracting indicators for distinguishing groups with poor progress in a collaborative process by analyzing the mechanism of interactive activation. Coding and statistical analysis are often adopted for analyzing such collaborative processes. In this study, using a multidimensional coding scheme with four dimensions designed to analyze the collaborative learning process more comprehensively and multilaterally, automatic coding is performed using deep learning methods and its accuracy is verified. In addition, this method is used to predict another dataset and determine its validity. The experiments of this study show that except for some typical misclassifications, the overall results were positive. However, in order to solve this misclassification problem, we also have to reconsider the redesign of the coding scheme and reexamine the classification technique by deep learning.