Using a combination of natural language and generative adversarial networks to predict time series and nonstructural data -- taking the rise and fall of TSMC's stock price as an example

Chien-Hung Lai, Yi Lin, Yuh-Shyan Hwang, Ya-Ting Chang

Abstract:

In recent years, with the rapid progress of artificial intelligence (AI) technology, the application of natural language processing (NLP) and generative adversarial network (GAN) has shown strong potential in the financial field. This research focuses on combining NLP and GAN technology to predict trends of the stock market. Taking Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC) as a case study, it explores the role of multi-modal data (including financial news sentiment, historical prices and technical indicators) in price rise and fall predictions. Bidirectional Encoder Representations from Transformers (BERT) model has been introduced to accurately classify financial
news sentiment (positive, neutral, negative), and combined with technical indicators, historical price data as input to the GAN to
generate future price trends. This research not only verifies the value of NLP technology in extracting time series data features, but also innovatively applies GAN to capture complex patterns of stock price fluctuations. At the end, experimental results show the achievement performs significantly better than traditional methods relied on indicators such as mean absolute percentage error (MAPE) and root mean square error (RMSE), providing a new perspective for financial data analysis and time series forecasting in Taiwan. Finally, the up to 5-day trend prediction performed well, with precision is over 94%.