Sentiment analysis with natural language processing methods on Turkish social media data

Several social media websites are showed up as Internet's improving continously and becoming an irreplaceable part of our lives. Those sites that people share their opinions and interact with others have become the address of scientists in terms of data source. People can access any information they need easily by doing research on Internet these days. Many of the data are open for feedbacks and these feedbacks are gathered for analyses of new ideas by surveys and forum sites. It is too hard to analyze feedbacks by a person as there are so many Internet users. At this point, emotion analysis concept showed up. Emotion analysis is aimed at classify the emotion as positive and negative by analyze the emotion and thought about a topic in texts. Entity property selection is used frequently nowadays in order to increase the performance and success in classification. Different methods are used in this selection and it is selecting the important qualities by eliminating the irrelevant features that affect the success in classification in target data set. Thus, hit ratio may increase. In this thesis, feature selection from Turkish texts written as colloquial is focused and support vector machine, artificial neural networks and centroid based classification algorithms are used on data that has detailed preprocessing. Gini Index, Information Gain and Genetic Algorithm are used as hybrids with 3 different classification algorithms on tweets belonging to 3 different GSM operators' followers. 100% success is achieved for 3 different GSM operators when genetic algorithm, which works as intuitively and has an important role in dimension reduction, and support vector machines are used hybridly.