Share market forecasting with LSTM neural network and sentimental trend prediction

Abstract

Forecasting and predicting future trend is getting signi cant importance in stock market exponently as it is volatile in nature. Stock market is an extremely complicated , unstable and volatile place due to the fact that prediction is very di cult. Because of uncertainty and having scope of gaining nancial pro ts, share market estimating and prediction has been a renowned matter in nancial and academic studies. Advanced algorithms of machine learning is required as there is no persistently appropriate prediction tool. Many research works from various sector have been done to overcome this di culties of predicting stock market. In machine learning sector a lot of research work already accomplished to predict share market. Many algorithms of Machine Learning have been utilized for this kind of prediction and the result was also satisfactory. In this thesis, we will extract all the real data from Dhaka Stock Exchange (DSE) using web scrapping and try to predict stock market price on a giving day, by approaching Long Short Term Memory(LSTM) Networks based on historical data mining method. The results of this paper show that Long Short Term Memory Networks can be applied for evaluation of historical stock pricing data and acquire valuable information by forecasting future trend with suitable nancial indicators. Beside this, we will extract all the news opinions from the respective web pages (DSE, Lonkabangla nancial port) and went through noise reduction, implementing algorithm and classi er to determine the sentiment polarity to come to a choice whether the stock price of a company are getting upward or downward trend. We are using na ve bayes classi er to examine the ratio of a sentence or phrase which can contain sentiment in from of positive, negative and neutral words. Using this model we can represent a status of some stock news.