Business analytics with data mining: An investigation of web based data with sentiment analysis

Consumers refer to product reviews in the processes of decision-making and obtaining information about the product before performing purchasing behaviour through e-commerce. Product reviews are produced by other consumers who have already purchased and experienced the product. It is aimed in this study to examine cosmetic products in a Turkey based e-commerce website with sentiment analysis and to create a new domain-specific Turkish sentiment dictionary model with manual labelling. In the study, a Turkish sentiment dictionary consisting of 65,378 words was created by manually labelling 875,445 product comments obtained from the web and sentiment analysis was performed using this dictionary. The data set is used for positive, neutral and negative classification problems by using various machine learning algorithms. Algorithms are compared by evaluated with accuracy, precision, recall and f-1 score metrics. The performance of the algorithms was highly successful in the groups and categories to which the product reviews were assigned. Compared to other algorithms, SVM showed the highest success in all categories. Thus, the created sentiment analysis dictionary showed classification success in the field of cosmetics and achieved high performance. The dictionary created in the study for the cosmetics sector is a reference source for similar or further studies to be carried out in the future.