Abstract:

Unlike classical sentiment analysis methods, Aspect-Based Sentiment Analysis (ABSA) can demonstrate a more successful performance in evaluating complex online consumer feedbacks including more than one category. As a matter of fact, consumer feedbacks on a platform can be referred to more than one aspect regarding a product, and standard sentiment analysis method is insufficient to analyse these comments. When the developments in the literature are reviewed, it is understood that HDTA studies are very popular among other studies focusing on sentiment analysis. In the SemEval ABSA-2016 competition, datasets were published in 8 different languages for HTDA and the teams competed for sentiment analysis. There are different subtasks in the competition, determining sub-categories such as aspect term, category and sentiment class. One of these subtasks is to determine the aspect term. HTDA studies for Turkish language are quite limited. There are studies using different languages and different word representation methods. There is no study examining the effect of word representation methods for the Turkish data set of SemEval Absa 2016 competition. This study was carried out to examine the success of different word representation methods in identifying aspect terms in customer comments. This study was carried out with the aim of examining the success of different word representation methods in identifying target terms in customer comments. Word2Vec, Glove and Fasttext word representation methods were examined within the scope of the analysis and it was seen that the method that could detect the aspect term most successfully was the Fasttext word representation method. The highest classification success for Turkish dataset in the literature with a success rate of 77% in terms of the F-1 score was also achieved in the study.

Keywords: