Stock Market (BSE) Prediction Using Unsupervised Sentiment Analysis and LSTM: A Hybrid Approach

Abstract

AbstractStock market prediction is an important factor for any Investor for gaining a maximum return. But, due to its volatility and uncertainty, the price prediction is challenging for Investors. The prediction of the Stock market mainly depends on certain parameters and aspects like social media data, country economy, financial news, historical prices, natural disasters, industry trends, and so on. In our research, we merged financial news data and historical prices for predicting nearby prices of the stock market. For this we collected financial news data from various sources related to it; pre-process them for extracting relevant key features and distributing them into various clusters according to their possible impact on stock market price. After combining these clusters with historical price data, we used Long-Short-Term Memory (LSTM) to predict the closing price. The advantage of using LSTM is that, it analyzes the relationship between sentiments and historic prices by storing past information that is important. This approach gives a more accurate model than the existing price prediction model using LSTM. Combining the sentiment cluster data with historical price data fills the gap in the existing approach of price prediction using only historical price data or using only news sentiment data.KeywordsUnsupervised sentiment analysisStock market predictionLong-short-term-memoryNatural language processingTerm frequency-inverse document frequency