A deep matching model for detecting reviews mismatched with products in e-commerce

Reviews of products in e-commerce are great value to both firms and customers. However, these reviews are not always genuine. Fake reviews can stymie firms and customers. Many studies have focused on detecting fake reviews. This paper addresses another kind of fake review on e-commerce websites called mismatched reviews where the content of reviews seems to be genuine, but they actually do not match the reviewed products. These reviews can harm the user experience in online shopping and damage the reputation of e-commerce websites. This paper develops a deep matching network to detect mismatched reviews, called MIRD, to calculate matching scores between the content of reviews and the reviewed products. MIRD encodes both reviews and products to the representation vectors and subsequently calculates a score in a latent space to measure how well the two vectors match. Experiments conducted on Yelp and Amazon datasets show that MIRD can recognize mismatched reviews well. Compared with the baseline models, MIRD obtains an 8% performance improvement on the two datasets. The experimental results demonstrate that our efforts to detect mismatched reviews are feasible and effective.