Title: PREDICTION OF ISTANBUL STOCK EXCHANGE (ISE) DIRECTION BASED ON NEWS ARTICLES

Year of Publication: 2013
Page Numbers: 320-330
Authors: Hakan Gunduz, Zehra Cataltepe
Conference Name: The Third International Conference on Digital Information Processing and Communications (ICDIPC2013)
- United Arab Emirates

Abstract:


In this paper, we examined the effects of financial news on Istanbul Stock Exchange and we tried to predict the direction of ISE National 100 Index open price after the news articles were published. In order to do this study, we got news articles from three big financial websites and we represented them as feature vectors. We obtained class labels from the ISE National 100 Index open price and assigned them to these feature vectors. After creating the datasets, we selected the informative features using Mutual Information (MI) and Term Frequency-Inverse Document Frequency (TF-IDF) Weighting methods. According to the selected features, we created feature subsets with different number of features and we trained a Naïve Bayes Classifier with them. We used precision, recall, accuracy and F-Measure metrics to evaluate the performances of our methods. We found out that feature selection can help us use a smaller number of features for classification and as a measure of how useful a feature is, TF-IDF is better than MI.