A Method for the Detection of Fake Reviews based on Temporal Features of Reviews and Comments

Abstract

Online reviews and comments have become an important resource for various decision making processes, such as sale and buy decisions. The truthfulness of online reviews is thus critical for both buyers and sellers since fake reviews will affect customer’s decisions due to misleading description and deceptive selling. This can cause financial loses for innocent customers. Fake review detection has thus attracted a lot of attention. However, most shopping websites have only focused on dealing with problematic reviews and comments. In this paper, we propose a method for the detection of outlier reviews based on reviewing records associated with products instead of just the reviews and comments. We first analyze the characteristics of such data using a crawled Amazon China dataset, revealing that the reviewing records of each product is similar for normal products. In the proposed method, we first extract the reviewing records of products to a temporal feature vector. We then develop an isolation forest algorithm to detect the outlier reviews of products based on the reviewing records of reviews and comments. We will verify the effectiveness of our proposed method and compare it to some existing temporal outlier detection methods using the crawled Amazon China dataset. We will also study the impact caused by the parameter selection of the reviewing records.