Classifying product review quality based on semantic and structural features

Abstract

Product reviews are written opinions submitted by consumers in assessing a product. The existence of product reviews is important because it can help consumers make better product purchasing decisions. But product reviews can also be unimportant if the quality of the information from the reviews is not helpful. This can be minimized if a classification is carried out to find out which reviews are helpful or not. For this to be achieved, this research will apply a support vector machine model using semantic and structural features to be able to classify review texts based on their characteristics. By applying the appropriate preprocessing stages, the final results show that the semantic features produce the highest F1-score value of 0.825. Whereas the structural features produce the highest F1-score value of 0.823. From this, it can be concluded that semantic features can be used to identify the characteristics of a review text that are helpful or not properly. This success also shows outstanding performance in classifying reviews as helpful or not compared to previous studies.