A Pattern Based Anti-Fraud Method in C2C Ecommerce Environment
With the growing popularity of online trading
sites, reputation systems are increasingly becoming an integral part of C2C
ecommerce systems. Reputation systems can collect, aggregate and distribute
participant feedback from past actions to encourage sellers' honest behaviors,
and effectively avoid cheating behaviors of those dishonest sellers. In such a
situation that neither buyers nor sellers are well informed of each other, the
reputation system is able to help buyers determine which sellers are more
credible. Such as eBay and Taobao[1][2][3][4], they all have their own
reputation systems. The world's largest C2C online auction site eBay has a reputation
system dealing with feedback information. Upon the completion of each
transaction, buyers and sellers have rights to give an rating points(-1, 0,
1)of the other[5]. Each participant will have an identification name, and its
evaluation will be given in connection with the transaction name on it.
Nowadays, many trading sites are using reputation systems like eBay's, while
some of them provide 1-5 rating range or use some other rating scales. Some of
them calculate the average feedback rating points while others calculate the
cumulative ones. These reputation-rating mechanisms can’t well deal with
thereputation slander, the reputation speculation and other means of fraud
generally. This leads to the reputation values given by reputation systems can’t
effectively reflect the performance of sellers, eventually leading to the
average benefit of buyers greatly reduced. In order to deal with the fraud
patterns mentioned above, Based on TRUST[8] model, we proposed a new fraud
pattern identification and filtering method. It is to find fraud pattern in
Time Window Scope and filter out those fraud ratings, Such as plenty of newer
buyers give higher ratings over threshold or lower ratings below threshold to a
fixed number of sellers, higher ratings over threshold are given by a fixed
number of sellers each other, etc. In this way, the reputation value that the
buyer computed will show much more fully of the true reputation of the sellers.
The experiment results in multi-agent system JADE prove that the method proposed
by us can make the sellers get more profit. The organization of the paper is as
follows. The second part introduces two kinds of fraud patterns that are very
regular and very hard to be recognized in ecommerce system. The third part
expounds the anti-fraud method we put forward based on TRUST; part four
illuminates the simulation experiment which is based on multi-agent system
JADE; part five is the summary of the paper and discusses the next step of our
study.
No comments:
Post a Comment