Akademik Veri Yönetim Sistemi
24th Signal Processing and Communication Application Conference (SIU), Zonguldak, Türkiye, 16 - 19 Mayıs 2016, ss.497-500
Sentiment analysis refers to classify the emotion of a text whether positive or negative. The studies conducted on sentiment analysis are generally based on English and other languages while there are limited studies on Turkish. In this study, after constructing a dataset using a well-known hotel reservation site booking.com, we compare the performances of different machine learning approaches. We also apply dictionary-based method, SentiTFIDF, which differs from the traditional methods due to their logarithmic differential term frequency and term presence distribution usage. The results are evaluated using the area under of a Receiver Operating Characteristic (ROC) curve (AUC). The results show that, using document term matrix as input gives better classification results than TFIDF matrix. We also observe that the best results are obtained using Random Forest classifier with an AUC value of % 89 on both positive and negative comments.