Effect of user's attribute on emotion estimation from Twitter
Kazuyuki Matsumoto, Minoru Yoshida, Kenji Kita, Yunong Wu, Fumihiro Iwasa · 2016
This study focused on emotion estimation from utterances on Twitter and analyzed the differences caused for estimation according to the attributes of users. SNS users possess various user attributes. If we could clarify the tendency for each attribute to involve emotion estimation, we would be able to create an emotion estimation model suitable for each attribute. This study considered two attributes, sex and job, and investigated the tendency of emotion estimation based on an emotion estimation model constructed by machine learning. In this paper, as a feature for machine learning, we used a sentence vector that was obtained by the summation of the word distributed expressions. As a result of the evaluation experiment using a method based on the nearest neighbor method, we obtained 0.4 higher estimation accuracies than using a baseline method based on a simple Bag of Words model.