Comparison of supervised and dictionary based sentiment analysis approaches on Turkish text

The more the World Wide Web (WWW) gets interactive, the more users share their opinions on it. Users share their opinions on products, services, brands, companies, news, etc. They share their opinions by using tools and technology, such as personal blogs, social networks like Facebook and Twitter, online newspapers, and e-commerce sites. This situation brings about an explosion of opinions. For example, it is not possible anymore to read all opinions shared on Web about a product, for both the company that sells the product and for users that buy it. Sentiment analysis is a research area for solving these kinds of problems. Sentiment Analysis encompasses operations such as determining the opinion bearing parts of a text, classifying the text by its sentiment orientation, presenting the opinions to users in easily understandable summarizations. In this work, two frequently used approaches by current research on Sentiment Analysis are compared. These approaches are sentiment analysis by using machine learning techniques and dictionary based sentiment analysis. Machine learning techniques are supervised techniques because they are based on learning from labeled training data. Dictionary based approach on the other hand, is a semi-supervised approach which starts from a seed sentiment vocabulary and extends this vocabulary by means of a semantic database such as WordNet to a sentiment dictionary and uses this dictionary as a resource for viisentiment classification tasks. In this work, specified approaches are applied toa Turkish dataset and results are discussed.Key Words : sentiment analysis, opinion mining, opinion extraction