Sentiment Analysis in Turkish Text with Machine Learning Algorithms

With the developing technology, the number of comments made on the internet is increasing day by day. It has become almost impossible to make a manual sentiment analysis on these comments. Therefore, new algorithms should be developed to automatically perform sentiment analysis on these texts. In this study, a sentiment analysis model has been developed for Turkish texts. While developing this model, lexicon-based methods and machine learning algorithms were used together. As a naive method of sentiment analysis, the root of each word in a sentence takes a score from a dictionary and the final polarity score of the relevant sentence is calculated by using additive score-based models. Machine learning models are trained to perform accurate sentiment annotations by using features based on polarity scores of texts. The final supervised machine learning model can achieve sentiment annotations of new Turkish texts within a 73% success rate without any human intervention.