Abstract:

With the development of technology, the e-commerce sector has grown very rapidly. Thanks to this development, a large increase is observed in the comments made on the products. As the number of reviews of the products increased, it became very difficult to examine and analyze them one by one, and the sellers became unable to evaluate all the comments made on their products. Within the scope of this study, a sentiment analysis study was conducted on e-commerce product reviews. First, 15170 product reviews were collected from e-commerce platforms operating in Turkey. In the labeling phase of the raw data set, the comments were labeled twice as positive and negative at different times and a two-class data set was prepared. With this dataset, a model was trained using the Long-Short-Term Memory algorithm, and a data set containing 203274 comments collected from e-commerce platforms was automatically labeled with this model. In the second stage of the study, the performance of the automatically labeled dataset was compared with the manually labeled dataset using four different deep learning algorithms (Recurrent Neural Networks, Long-Short-Term Memory, Gated Recurrent Unit, Convolutional Neural Networks). In the last stage of the study, classification performances of four traditional machine learning (Logistic Regression, Support Vector Machines, Random Forest, Naive Bayes) and four different deep learning algorithms were compared using automatically labeled datasets with high classification performance. As a result of the experimental studies, it has been observed that deep learning algorithms are more successful than machine learning algorithms, and high results have been obtained in experiments where the data set automatically labeled with the developed model is used.

Keywords: