Improvising the Stock Prediction by Integrating with roBERTa and LSTM

Abstract

In the Stock market, the volatility of leading MNCs’(Multi-National Corporations) shares is a major matter of concern and comes under the limelight nowadays. Unlike the 1920s sudden surge and dot-com crash, the contemporary world has never seen such a biggest bull or bear particularly in the past few decades. The stock market is majorly influenced by the credibility opinion of the general public on the firm. In the 21st century, the emergence of research LLC (Limited Liability Company) which gains profit from short selling of the shares by manipulating the share of a certain firm by exposing the legality of trespassing norms has made the researchers include a current public sentiment on the firm since short selling is a matter of one day. The first and foremost impact of such exposure would be instantly taken to Twitter, a credible social media. In order to infer the associativity of sentiment analysis on the stock market analysis we have taken time-series data of a recently exposed firm which faces the biggest bear in the market from Yahoo Finance for the timeline of 07-02-22 to 03-02-2023 and the Twitter data for the same timeline had been accessed by is a scraper for Social Networking Services (SNS). The extracted tweet data with almost 1000 tweets each day has been analyzed by Meta’s roBERTa, an NLP(Natural Language Processing)-based framework for sentiment analysis. It is used to predict whether the market will be bearish or bullish on the day. Then the sentiment flag attribute and the market data attribute have been used to build a 3-layered Long Short Term Memory (LSTM), an ANN(Artificial Neural Network) where the data will be predicted for the same day’s stock movement. The results show that the sentiment reflects on the stock’s movement and the accuracy of the proposed work is about 96.14%.