Power Harassment Judgement Method Based on Emotion Analysis Using Natural Language Processing

Abstract

In recent years, the number of Power Harassment cases in Japan has been increasing. Power Harassment is defined as behavior that meets all three of the following criteria: first, it is behavior that is based on a superior relationship; second, it is behavior that is unnecessary and beyond the reasonable scope of work; and third, it is behavior that is detrimental to the work environment. The purpose of this study is to notify the perpetrator when the possibility of Power Harassment is judged to be high based on the conversational voice. As a preliminary step toward this goal, we experimented to determine whether the target text constitutes Power Harassment or not using emotion analysis by natural language processing. The system for sentiment analysis in Japanese was built based on Mecab and CaboCha. The polarity dictionary required for the sentiment analysis was created using the FastText natural language processing model based on text data from past cases of Power Harassment. As a result, a discrimination rate of about 84% was obtained. This is a better result than previous studies. If this system is put into practical use, it could reduce Power Harassment, which has become a social problem in Japan