An Adaptive Graph Cut Algorithm for Spammer Group Detection from Weighted One Mode Projection of Bipartite Graph

Abstract

Now-a-days online reviews have become an indispensable resource for customers decision making before purchasing any product. Product reviews provide valuable information that helps customers, retailers, and product owners in making decision to buy or sell product online. The rise of ecommerce platforms has led to an increase volume of customer review data available in the internet. Though most of the customers write review about the product solidness, quality, usability or else but it has also been also found that some people write intentional collusive review to mislead customer's decision-making process and damage the online business as well. These intentional fake reviews are written in order to rise or down a business and they are called spam reviews. The persons who writes fake review are called spammer. Sometimes review spammer works in group to control the overall sentiment or notion about a product, item or a business. When the review spammer works in collaboration they are called group spammer. Most of the research works found in the literature worked on individual spammer detection using the traditional machine learning or neural network-based algorithms. Very few researchers worked on group spammer detection. This paper focuses on group spammer detection from online reviews and proposed a weighted one-mode bipartite projection graph-based algorithm that extract the structural, behavioral and time-centric features of reviews to detect group spammer from online marketplace.