Identifying Hijacked Reviews
Abstract
Customers on online marketplaces have to proceed with extreme caution before buying any product as they cannot evaluate it physically. Reviews are crucial metrics to gauge the quality and authenticity of the item. This dependence on reviews has led to a rise in the number of unethical sellers who exploit the review system in e-commerce websites via fraudulent techniques, like fake reviews.
Fake reviews have been an actively researched domain for the last decade as an independent review problem and a behavior pattern recognition problem. While almost everyone is looking around to detect fake reviews, we are looking at a different facet of e-commerce fraud which is called “Review Hijacking” (or “Review reuse” or “Bait-and-Switch review”).
Review hijacking is a new review manipulation tactic in which black-hat sellers “hijack” existing review listings of a product and use them to sell their products with no reviews. These items may be discontinued and unrelated but contain many positive reviews. More favorable ratings lead to better search ranking, make a new product appear well-reviewed and legitimate, and, ultimately, boost sales.
There has been little academic research for this review scam. Hence, we introduce what review hijacking is, the methods used to employ it, challenges to identify, and the impact it has caused. We further find techniques to uncover such cases.
We analyze the extent of this problem by applying various Information Retrieval methods like Boolean Retrieval (BIR), TF-IDF and Topic modeling on Amazon public datasets. Then, we synthetically label our data using Weak Supervision and by swapping the product-review pairs to run supervised learning models. We employ Deep Learning methods like Siamese LSTM and BERT Sentence Pair Classification to detect this e-commerce fraud efficiently on a larger scale.
Subject
Review fraudAmazon reviews
NLP
BERT
Review Hijacking
Synthetic data labeling
Information Retrieval
Siamese network
LSTM
Natural Language Processing
Machine Learning
Citation
Daryani, Monika Manohar (2021). Identifying Hijacked Reviews. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /195772.