Stock Market Prediction using Reinforcement Learning with Sentiment Analysis


Xuemei Li and Hua Ming, Oakland University, USA


This work creates a new Deep Q-learning model with augmented sentiment analysis and stock trend labelling (DQS model). The novelty of this study is as following. We form the stock price prediction problem as trend prediction instead of predicting its accurate price. By benchmarking multiple machine learning methods, stock market trend label is proven to be effective and can be predicted accurately. We use news titles and apply Valence Aware Dictionary for Sentiment Reasoning (VADER) to project the sentiment of the news about stock under study. The input feature to a customized Deep Q-learning model incorporates stock market trend label and sentiment analysis score label. Our study shows that a trading agent using DQS model achieved 83% more portfolio value than a DQ model using only stock technical indicators. The trading agent based on DQS model achieved a Sharpe ratio of 3.65 comparing with 1.6 achieved by a traditional DQ model-based trading agent. This indicates the DQS model combining with input features proposed by our study can achieve excellent risk-free investment portfolio.


Machine Learning, Deep Q-learning, Sentiment Analysis, Stock Market Prediction.