Topic estimation during group discussion using Topic Model

Abstract

In our laboratory, we are conducting research to analyze the characteristics of infants based on videos of group discussions of infants with the aim of reducing the burden on early childhood education for nursery teachers. In the processing before analysis, the numerical conversion of voice / action information from the moving image is performed manually. This process takes time to create data, and the evaluation is subjective, so an objective evaluation is required. Therefore, it is necessary to automate this process. In order to automate of numerical conversion of voice / behavior information from video, the purpose of this research is to acquire voice information such as "Are you talking about the theme?" The topic of the group was estimated by using LDA by dividing the video of the discussion at regular intervals, and the effectiveness of the method was verified.