A Scalable Approach for Sentiment Analysis of Turkish Tweets and Linking Tweets To News

Abstract

We present a framework for sentiment analysis on tweets related to news items. Given a set of tweets and news items, our framework classifies tweets as positive or negative and links them to the related news items. For the classification of tweets we use three of the most used machine learning methods, namely Naive Bayes, Complementary Naive Bayes, and Logistic Regression, and for linking tweets to news items, Natural Language Processing (NLP) techniques are used, including Zemberek NLP library for stemming and morphological analysis and then bag-of-words method for mapping. To test the framework, we collected 6000 tweets and labeled them manually to build a classifier for sentiment analysis. We considered tweets and news in Turkish language only in this work. Our results show that Naive Bayes performs well on classifying tweets in Turkish.