E-Ticaret Sitelerinin Tüketici Yorumlarında Manipülasyon Olup Olmadığının Benford Analizi ile Yorumlanması
Özet
Today, online shopping through e-commerce sites is increasing very rapidly. While making the purchasing decision during the transactions made by the consumers on the e-commerce sites, the online reviews left before for the relevant store or product are effective. Due to this effect of online reviews, it is seen that customers are tried to be manipulated from time to time through e-commerce sites by using fraudulent or fake online reviews, in order to make the product/store look better/worse than it is.
The aim of this study is to determine whether there is a manipulation effect or fraudulent situation in online reviews on a particular e-commerce site by using Benford analysis. In this study, the number of online reviews, word counts of online reviews and character counts of online reviews available on e-commerce sites are analyzed by using Benford analysis. In addition, the analysis of the number of positive, negative, neutral and total reviews was carried out with Benford's law. Benford analysis, which is used in fraud auditing in tax and social media audit applications as well as in criminal research, suggests that the distribution of numerical values in data formed by natural methods is within a pattern. Deviation of the found distribution from the expected distribution is considered a sign of systematic manipulation (or unnatural interventions).
According to the Benford analysis results using the mean absolute deviation test of the online reviews made by the customers on the PazarYeri.xyz e-commerce site between 01/05/2020 and 01/05/2021, the results of the Benford analysis for the number of reviews and the word count of the reviews were found to be inconsistent; which means that there is a manipulation sign for the data. For the character counts of the reviews, the Benford analysis result was determined as acceptable compatibility. The fact that the number of characters in the online reviews made by the users is in compliance with the law can be interpreted as there is no manipulation in the reviews. In this case, it is thought that customer reviews are added or removed as a whole, and therefore, although a consistent result is obtained in the number of characters, there is an inconsistency in the number of reviews and word counts. The distortion factor value is negative in the results of the inconsistent review count and the word count of the online reviews. When the distortion factor values of the character count analysis of the reviews, which come out as acceptable compatibility, are compared with these values, it is understood that they are considerably high. It is understood that there is less data than it should be due to the negative distortion factor. In other words, it is thought that some reviews were not published due to the filters and rules set by the e-commerce site.
Within the method used in this study, it has been determined that the number of online reviews made on the e-commerce site, the number of words in the reviews and the number of characters in the reviews are the variables that need to be analyzed together and that they should be analyzed separately for positive, negative and neutral comments. Besides, it has been revealed that it is important to consider the Distortion Factor values in the interpretations when performing the Benford analysis with the Mean Absolute Deviation Test.