Sentiments Analysis Prediction of The Arabic Stock-Market News Based on Machine- and Deep-Learning Approaches

Abstract

Stock market prediction of companies is a vital interest for financial analysts, investors, and other competitors. There is difficulty in predicting the future status of the companies' stocks. However, stock market behavior depends on the polarity prediction classification. Therefore, it is essential to use sentiment analysis to study attention indicators for stock market behavior in the news. Sentiment analysis (SA) can be used to extract public sentiments from stock news microblog platforms. Previous studies used machine learning (ML) algorithms to classify Arabic stock news into positive, negative, or neutral types. Recently, deep learning (DL) algorithms have provided good accuracy for Arabic SA. Motivated by such results; this study applies ML and DL techniques to classify sentiments of Arabic stock news. 30,098 articles were collected and preprocessed from the Saudi stock-market platform, Tadawul. For the sake of comparison, two ML and two DL techniques were performed for SA: Naive Bayes (NB), logistic regression, fast-text, and long short-term memory (LSTM). These algorithms were used to classify the sentiments of the collected data and help investors and stock analysts with decision-making. The results show that the DL techniques outperformed the ML algorithm. The experimental result of the LSTM model was 84%, which is the same as the reduced-features logistic regression model, but it has the lowest features over the same timeframe. Therefore, the LSTM model simultaneously has the best accuracy and the fewest features. On the other hand, the NB models achieve the worst performance. The Arabic SA models assist in decision-making based on the stocks news sentiments predicting the upcoming stock trends for the investors or analysts. These sentiment's models would limit the risks by supporting the decision-making analysts. Thus, this study would be a valuable resource to the stock market sector based on Arabic linguistic features.