Fraud Detection in E-Commerce Mohit Kumar/Disha Makhija
Aug 18, 2015
Fraud Detection in E-Commerce
Mohit Kumar/Disha Makhija
Types of Fraud in E-Commerce Marketplace
● Buyer fraud○ Credit Card Fraud○ Reseller Fraud○ COD/RTO Fraud○ Product Exchange Fraud
● Seller fraud○ Reviews/Ratings Fraud○ Fake Listing○ Price Abuse (MRP abuse)○ Brand Infringement○ Seller Protection Fund Fraud
Reviews and Ratings Fraud Detection in E-Commerce
Disha MakhijaOngoing collaboration with: Prof Christos Faloutsos’s group (CMU)
Product Ratings, Reviews and Upvotes
Motivation for Product Ratings Fraud: Conversion vs Rating
Lower rated products have
conversion drop while higher
rated products have
conversion increase:
• Direct financial motivation
for sellers/publishers to
boost their products
Seller Ratings
1 product, multiple sellers,comparable prices
Who would you buy from???
Seller Rating Fraud - Simple Pattern
Seller Rating Fraud: Boosting Self Pulling Competition
Product Review Fraud Detection: Boosting products
Review & Ratings Fraud Literature
2011 2013 20142012
Ott et al., ACL
Bing et al., ICDM
Bing et al., EMNLP,
Christos et al., AAAI,
Christos et al., KDD
Bing et al., ICDM
NLP based
ML based
Network AnalysisChristos et al., WWW
Bing et al., AAAI
Bing et al., KDD
A technique based on synchronous and abnormal behaviour : CatchSync
● Idea : users intending fraudulent activities tend to behave synchronously, which is not quite normal
CatchSync- Catching Synchronized Behavior in Large Directed Graphs, Faloutsos et al,KDD 2014
One user, different sellers but competition!!
11
A technique based on synchronous and abnormal behaviour : CatchSync
● Idea : users intending fraudulent activities tend to behave synchronously, which is not quite normal
CatchSync- Catching Synchronized Behavior in Large Directed Graphs, Faloutsos et al,KDD 2014
Precision for Top K Users: 48%
12
Review graph based online store spammer detection
Review graph based online store spammer detection, Bing et al, ICDM 2011 13
Review graph based online store spammer detection
Precision for Top K Users: 39%
Review graph based online store spammer detection, Bing et al, ICDM 2011 14
● Precision @ top 100 Users reviewed - 63%○ Using ensemble of state-of-art techniques○ Best state of the art method individually gives 48
Current Detection Performance
15
Summary
Summary - Sold my kidney!
Thanks!