Towards Machine Evaluation of Human Produced Comments in Turkish

Migena Ceyhan, Zeynep Orhan and Dimitrios Karras
University of Epoka

Abstract

The digital era has significantly altered the traditional human relationships. Opinions, emotions, and ideas are continually being shared. If these data out of various fields of life can be used and processed using different applications, significant outcomes can be obtained. Each passing day the number of such kind of applications is increasing, proving the benefits achieved by social media text analysis. In the current paper movie reviews picked from social media are categorized into positive or negative by performing a sentiment analysis of their linguistic characteristic features. The key point of this investigation is to characterize new reviews automatically. This is an important process in the automatization of information classification. To start with, all the unique words roots and their usage frequencies from each class of user comments are processed with different parameters to form features sets or subsets, which are used to train the system according to three machine learning algorithms. The paper compares the results to get a better insight of the feature selection task and its importance. The high accuracies from some of machine learning – feature selection combinations are evaluated motivating for important future research. Keywords: Sentiment Analysis, Turkish Language Processing, movie reviews, Machine Learning, feature refining.


Download Proceedings Book Volume1
Download Proceedings Book Volume2

Presentation