Understanding Large-Scale Network Effects in Detecting Review Spammers

Thumbnail Image

Files

Understanding_Large-Scale_Network_Effects_in_Detecting_Review_Spammers.pdf (5.66 MB)

Links to Files

https://ieeexplore.ieee.org/abstract/document/10044638

Permanent Link

https://doi.org/10.1109/TCSS.2023.3243139
 http://hdl.handle.net/11603/27026

Collections

UMBC Information Systems Department
UMBC Faculty Collection

Author/Creator

Author/Creator ORCID

https://orcid.org/0000-0003-2631-9223

Date

2023-02-14

Type of Work

journal articles
postprints
Text

Department

Program

Citation of Original Publication

J. K. Rout, K. S. Sahoo, A. Dalmia, S. Bakshi, M. Bilal and H. Song, "Understanding Large-Scale Network Effects in Detecting Review Spammers," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2023.3243139.

Rights

© 2023 IEEE.  Personal use of this material is permitted.  Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Subjects

Abstract

Opinion spam detection is a challenge for online review systems and social forum operators. Opinion spamming costs businesses and people money since it deceives customers as well as automated opinion mining and sentiment analysis systems by bestowing undeserved positive opinions on target firms and/or bestowing fake negative opinions on others. One popular detection approach is to model a review system as a network of users, products, and reviews, for example using review graph models. In this article, we study the effects of network scale on network-based review spammer detection models, specifically on the trust model and the SpammerRank model. We then evaluate both network models using two large publicly available review datasets, namely: the Amazon dataset (containing 6 million reviews by more than 2 million reviewers) and the UCSD dataset (containing over 82 million reviews by 21 million reviewers). It has been observed thatSpammerRank model provides a better scaling time for applications requiring reviewer indicators and in case of trust model distributions are flattening out indicating variance of reviews with respect to spamming. Detailed observations on the scaling effects of these models are reported in the result section