Sentiment analysis in turkish social media texts

In this study, a method for sentiment analysis of Turkish social media texts is proposed. Sentiment analysis was performed on the Twitter text data using the emotion categories determined by the field study. Social media texts were classified into 10 emotion categories as "Happy", "Trust", "Appreciation", "Pride", "Expectation", "Recommendation", "Curiosity", "Disappointment" and the most effective words for each emotion are determined. Emotions used in this study are based on R. Plutchik's emotion theory. The text classification methods for sentiment analysis used in the study are Naïve Bayes, Decision Trees, K-Nearest Neighbors and Support Vector Machines. According to the results, correct classification performance of Appreciation is 65%. After completing the classification process, the forward selection method was used to find the most important words expressing the emotion. The most effective words for 10 emotions were determined using the results. Appreciation is the most successful emotion class among the other emotions considering the words found in the process. The most effective words for appreciation are "teşekkürler, helal, teşekkür, tebrikler, adamsın, bravo, davranış". In this study, Twitter messages posted to formal account of a GSM company is analyzed. According to the analysis results, it has been shown that emotional states and subjects of the tweets could be determined by the words of user tweets.