Semanticaly polarizing data from a social network

In this thesis, it is aimed to make dirty and irregular information circulating on the internet more convenient by taking advantage of the tweets that twitter users have taken. Increasing use of social networking today and the intensive demand of users have brought about the control of shared content. Prevention of malicious people 's perception management, provocation, fraudulent texts reveals the idea of classifying existing data. The notion of classification is that the ethical values of the community are to be semantically positive, negative, neutral, taking into account language and dialect. Therefore, NodeXL for analysis and data received via Twitter are eliminated from unnecessary words and punctuation, misspellings, polish differences have been eliminated. After the configuration phase, the tweet was tagged as positive, negative, neutral by five different volunteers each tweet in order to get more accurate results, and the tag value with high number was accepted as the result. The prepared data set is created by feature vectors according to the word frequency used with the help of the knime program, and the learning and estimation groups of 70 to 30 are classified by SVM (Support Vevtor Machine), Decision Tree and Naive Bayes algorithms.