Opinion Spam Detection Based on Heterogeneous Information Network

Abstract

User-generated online reviews can play a significant role in the success of retail products, hotels, restaurants, etc. However, opinion spam has become a widespread problem because often paid spam reviewers write fake reviews to unjustly promote or demote certain products or businesses. Existing approaches often utilize part of the clues within user-review-product Heterogeneous Information Network to detect spam or spammers, which cannot get satisfied performance. In this work, we use novel features like posted photos in reviews and user social networks, and propose a new approach called SkyNet that utilizes clues from all heterogeneous data including metadata (text, photos within reviews, and etc.) as well as relational data, and harness them collectively under a unified framework to spot suspicious users and reviews. The experiments showed that SkyNet significantly outperforms several baselines and state-of-the-art methods on real-world review dataset.