Github Link

Abstract

The analysis of the stock market has been very profound among the machine learning community. It is well known that is it not very easy to predict the future stock price of a company. There are a variety of intrinsic and extrinsic factors that might influence the market. Many predictive approaches ranging from statistical models to advanced deep learning models have been developed. Most people use historical price data and other technical measures to analyze the stock movement. It is also interesting to see that the news about the company has some kind of influence over the choices of an investor. This paper presents and compares a range of prediction techniques that can be used to predict the stock market. To begin with, the stock data along with that the news of a stock company were extracted from Yahoo Finance and Twitter respectively, and then sentiment scores were extracted from the text using BERT (Bidirectional Encoder Representations from Transformers) which is a state-of-the-art language model. Following that, deep learning models namely LSTM (Long Short Term Memory) and GAN (Generative Adversarial Network) were built and these results were compared with a simpler ARIMAX (Auto-Regressive Integrated Moving Average) model.

Research paper