Language generation by GPT-2 and text analysis of economic fluctuations forecast through LSTM

Abstract

Economic fluctuations have a great impact on business management. In this study, we explorethe method to improve the accuracy of the prediction model of the economic fluctuation using the economicwatcher survey reported by the Cabinet Office, the Government of Japan. The economic watcher surveyincludes interviewees' answers in the text and the economic business trend tag for each answer. Theclassification was attempted through the LSTM model. As a result of the analysis, we found the predictionaccuracy could be improved by using texts generated by GPT-2. Further examination of the classificationmodel will be planned.