An Approach for Movie Review Classification in Turkish

Abstract

Web 2.0 has given to all people the right to become a representative of a huge cast of informal media. The importance of this power is getting more evident everyday. Every social media actor can influence the rest of the world by one’s own opinions, feelings, and thoughts generously shared on multiple media. This information belonging to various fields of life can be very handy and be used to one’s advantage, gaining precious experience. One of the greatest problems that this poses is the huge number of data spread everywhere, which are difficult to process as row data per se. Social media and general sentiment text analysis is of much valuable use, accomplishing the task extracting pure gold out of raw mineral. The key point of this investigation is to characterize new reviews automatically. To start with, features selected out of all the word roots appearing in the comments were used to train the system according to known machine learning algorithms. Next, critical words determining positive or negative sense were extracted. Another strategy was attempted eliminating common terms and dealing only with the significant class-determining words to build vocabulary with them. Aparts from linear approach, vector based feature sets were prepared out all or some of the features. The outcomes acquired were analyzed and compared leading to important conclusions, emphasizing the importance of feature selection in text classification.