A Bidirectional Encoder Representations from Transformers – Multi Head Attention for Predicting the Stock Market Volatility Based on Textual Sentiment Analysis

Recent days, stock market plays an important role in country's economic landscape, serving as a vital conduit for companies and investors to facilitate investments. As a cornerstone of economic growth. The stock market offers unparalleled opportunities for businesses to access capital and expand their operations. The existing approaches for stock market prediction based on textual sentiment analysis has gradient vanishing problems and limited parallelization. Therefore, this paper proposes Bidirectional Encoder Representations from Transformers (BERT)- Multi Head Attention which effectively enabled parallel processing and the transformer minimizes the gradient vanishing problem. Initially, the data is taken from Xueqiu investors community and the processing of text is done in three steps they are web crawling, data cleaning and data transforming. Then, this data is used for measuring volatility which helps the investors assess market risk and potential losses. Finally, proposed BERT- Multi Head Attention is used for predicting and effectively enabled the parallel processing and minimizes the problem of vanishing gradient. The proposed BERT- Multi Head Attention achieved better results in terms of accuracy of 90.65% and Fl-Score of 88.76% when compared with existing NB.