Bayesian social reinforcement for stock trend prediction

Abstract

Stock trend prediction has been a challenging and relevant task for both conventional machine learning and deep learning methods. To this end, multiple approaches have been developed in the literature with the application of machine learning, specifically sentiment analysis with natural language processing. However, the majority of finance-based machine learning research has been done with a deterministic approach rather than a probabilistic approach. Decision making within the stock market is challenging because of its inherent stochastic nature and volatility. In this paper, we propose a general framework for social reinforcement of public investment sentiments, before presenting both a na¨ıve and Bayesian approach for reinforcing sentiment scores by incorporating additional information from social media, to improve stock trend predictions. As a side product, the duration of the impacts of the sentiments and their social reinforcement on the stock trend is examined.