Incorporating Media Bias in Sentiment Analysis for Improved Stock Prediction Using Neural Networks

In this work we examine the effects of incorporating media bias and its impact on stock prediction. Sentiment analysis finance articles in the field give the model actual news data and from there it forms a sentiment score into an LSTM. There is no indication about what news source this came from and there is a flaw in this. There is media bias in this as certain news outlets want to push a certain narrative which could have no meaning on the actual prediction, we plan to incorporate this in the predictions in order to ensure that this media bias is accounted for and thus get more accurate predictions. With everything incorporated this model has a mean absolute percent error close to 1 percent.


In this work we examine the effects of incorporating media bias and its impact on stock prediction. We achieve results that are state of the art in academia for stock price prediction that include the use of sentiment analysis. We research and present findings of alternate techniques that can achieve similar results and finally incorporate media bias for the sentiment. The findings strongly suggest that sentiment analysis should incorporate the source of the information as not all sources are created equal in quality and in bias.