Sentiment Analysis of Japan Economic Watcher Survey Data with Japanese BERT model

Abstract

A new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers obtains new state-of-the-art results on eleven natural language processing tasks in English. We build a Japanese version of BERT model with Japanese Wikipedia data and perform sentiment analysis of Japan Economic Watcher Survey Data. We confirmed that the result of sentiment analysis using the Japanese version of BERT model is better than the result without the model.