Polarity detection of Turkish comments on technology companies

Abstract

In this study, comments about technology brands are collected from a popular Turkish website, eksisözlük, and classified as positive or negative. Turkish text is preprocessed with different kinds of filters and then modeled with 1-gram, 2-grams and 3-grams language models. Naive Bayes (NB), Support Vector Machines (SVM) and K nearest neighbor (KNN) classifiers are applied on different configurations of preprocessing techniques, language models and linguistic attributes for comparison. We measured best F-measure as 0,696 on our test dataset.