A Lexicon Features‐Based CNN Model for Stock Price Prediction

Abstract Stock price prediction is a complex and challenging process due to its highly fluctuating and volatile characteristics. Factors that contribute to such characteristics include political and economic situations, company's finances, budgetary news, and events related to public safety such as wars, public unrest, and so on. Sentimental analysis is a prominent approach that is frequently used in a variety of industries. These help machines in determining the emotion behind a text (sentence) as positive, negative, or neutral. The sentimental models are used in curating human language (English) and transforming them in model interpretable format. In this article, we attempt to predict stock market prices using sentiment analysis. We have used TATA Consulting Services (TCS), a global multinational technology company as being our selected stock in the stock market. News and stock price data were accumulated from the Economic Times newspaper and the National Stock Exchange of India, respectively. VADER (Valence Aware Dictionary and sEntiment Reasoner) sentiment model has been used to get separate polarity scores for the collected news titles and descriptions. A dataset has been built using the polarity scores and stock market price data for modelling a one‐dimensional convolutional neural network (CNN). The model predicts ‘High Value’, ‘Low Value’, ‘Open Value’, and ‘Prev. Close Value’. These were evaluated using appropriate evaluation metrics and the outcome or predictions were analysed in terms of a real‐world trader perspective and actual readings.