Try to Find Fake Reviews with Semantic and Relational Discovery

摘要

There are a large number of online fake reviews reviewed from imposters intended to promote or demote the reputation and sales of target items. These activities result in vicious competition and damage the interests of individuals. Several methods have been proposed from related researchers to detect the fake reviews, such as content based methods, behavioral based methods, graph based discovery and bursting pattern discovery. In this work, we focus on detecting fake reviews based on semantic and relational discovery. That is, we find that there are some differences in related topics of fake reviews and non-fake reviews. There are also some close relationships of reviews, reviewers and items. As such, we first extracted semantic based features and relational based features with several data mining techniques, then treated the fake review detection problem as binary classification task and built classification models. In the end, we demonstrated the validity of our method through our designed experiments.