Automated Analyzing System for Recognizing the Elemental Processes Based on the Labeled LDA

Abstract

In this paper, we described an automated analyzing method for the elemental processes. This method predicted the elemental processes from the sensor data by using labeled latent Dirichlet allocation (L-LDA) that is supervised topic model. The L-LDA studies automatically characteristic motion. We do not need to define characteristic motion by applying the L-LDA to motion analysis. The sensor data are motion sensors of both hands and a pressure sensor of working space. Numerical data obtained from the sensors were converted into word data by the threshold process using statistically determined thresholds. The automated analysis by the L-LDA was conducted by using the word data. We confirmed that recall by the method was over 86.9% by the evaluation experiment.