MODELING STOCK PRICE MOVEMENTS PREDICTION BASED ON NEWS SENTIMENT ANALYSIS AND DEEP LEARNING

Abstract

Nowadays, with the rapid growth of information spread, investors involve news and sentiments in their financial decision more than before. This paper investigates the effect of technical and fundamental analysis in the form of technical indicators and sentiments of news on Iranian stocks. Several packages and technologies are developed for English semantic; in this regard, most previous works are done on English, especially Twitter. On the other hand, there are rare attempts about the effect of Persian semantics on Iranian stocks due to the lack of uniform packages and technologies. This study collects news articles in Iran that are related to stocks. After data preprocessing, the polarity of news is discerned by the HESNEGAR lexicon. It is the first to consider a semantic Persian lexicon on Iranian stocks. Three models are proposed based on the deep learning approach-convolutional neural networks; price only, news sentiments and hybrid models. Experimental results showed that hybrid model considering both technical indicators and news sentiments using the HESNEGAR lexicon could significantly improve the prediction accuracy compared to price only and news sentiments models. This study can be the reference model to plan a trading strategy.