عنوان مقاله [English]
With the growth of electronic commerce, the number of customer reviews on e-commerce websites is growing. These reviews contain valuable information that helps future customers make better decisions about their purchases and allows retailers to promote their products, services and marketing solutions. These reviews have been a major target for fraudster attacks, as they directly influence customer purchase decisions. Fraudsters are usually deployed by companies to write fake reviews to promote their products and to divert customers from Competitors. Submitted reviews are typically done by experienced professionals, with the aim of writing plausible criticism. Due to the above mentioned issue and the fact that customer review mining without removing fake reviews will have low efficiency, fraud detection and automatic identification of fraudsters in online reviews is a necessity for e-commerce. This study designs a hybrid system to detect online fake reviews that uses the features of objects and the relationships between various entities simultaneously. Local features are the textual and non-textual features of reviews, the behavioral and demographic attributes of reviewers and product properties. The relations and classes of related entities are also used as predictors. Local features and network effects, between them, simultaneously reveal some parts of fraud signs, and development of a system to detect fake reviews is the main aim of this research. A communication network of reviews and users is formed based on the trust and block network between users, feedback for reviews, and common users and products. To achieve this goal, relational data mining with a relational dependency network algorithm is used that combines local features and relations internally. This method is performed on two review data sets and the results show the improvement in the efficiency of this approach in comparison to similar methods. In addition, this method provides a flexible method for diverse comment networks, based on the considered dataset,without the need to intuitive assumptions about the relationship between entities.