Time Frame Analysis for Sentiment Prediction of Stock Based on Financial News using Natural Language Processing

Abstract

This research is a study on the impact of a specific stock sentiment based on its news, previous stock movements and finally finding investors sentiment over the stock. This study leverages daily Indian financial news between 2017 and 2021, extracted from various Indian and foreign news sources such as Economic Times, Money Control, Livemint, Business Today, NY Times, WSJ and Washington Post. In this work we propose to analyze news data with a unique pre-processing method that uses vectorization and BERT data processing technology. This is followed by a comparative study and predictive machine learning analysis of following models - Naive Bayes and Recurrent Neural Networks (RNN) with Gated Recurrent Units (GRU), Bi-directional Long Short Term Memory (LSTM) and RNN-LSTM with the pre-processed news data leading us to better accuracy and sentiment findings as compared to other approaches. Based on the comparisons, the results show that - Bi-Directional LSTM layer based on RNN architecture along with BERT Data Processing gives an accuracy of 90.15% leading us to a conclusion of adding a layer of BERT data processing for sentiment analysis to get better results. Further an application feature is being proposed which analyzes real-time stock financial news using RNN-Bi-Directional LSTM, giving a confidence value that is used to calculate overall sentiment of a stock being traded in Indian Stock Exchange for different time frames.