Proposal of Emotion Tagging Method Using Emotion-Related Words and its Application to News Site Access Log Analysis

Abstract

An emotion tagging method using emotion-related words is proposed and applied to access log analysis of news site in this paper. We regard emotion as the feeling which ordinary people have from the viewpoint of a third party. The system tags one of 8 emotions (joy, like, safe, fear, anger, disgust, sad, shame) to one sentence. There are 5 steps in the proposed emotion tagging method. 1) The system tags emotion to the sentence directly which have the emotion-related word in emotion database. 2) The system does morphological analysis for the sentence which can't be tagged emotion in the first step. 3) Searching emotion queries based on emotion-related words are made using emotion database. 4) The system posts the searching emotion queries and each word in the sentence to Mainichi news database and calculates the co-occurrence score. 5) The system changes the score considering the importance of each word. The emotion which has the best co-occurrence score is tagged to the sentence. The experiment result shows that the rate of correctness of the proposed emotion tagging method is about 80%. For emotion based access log analysis, we gathered the access logs of Asahi web news site. The system tags emotion using proposed method to each gathered news headlines. After emotion tagging, the system extracts the emotion based rules from the access logs by counting the frequencies of emotion sequences. As a result of access log analysis we obtained some interesting rules such that users tend to read news which convey positive emotion successively.