Interesting article in today’s WSJ about Ebay and computer scientists who track fraud through algori

Longacre

I don’t have a link to the article in today's Wall Street Journal, but it is pretty interesting. I am no math geek, so bear with me as I try to explain the article. A group of computer scientists from Carnegie Mellon developed a theory called “bipartite core” in order to predict fraud on Ebay. A bipartite core is a term from math and computer science that describes a situation where members of Group A will deal with members of Group B, but never with each other; the pattern become apparent when the interactions of the two groups are plotted on a graph.

To my simple mind, an example of this is a fraudster creating multiple Ebay accounts and creating multiple accomplice accounts and generating positive feedback with related party transactions and a few outside people. The article explains this a lot better than I can, but supposedly these computer scientists created an algorithm that can analyze a large amount of data on Ebay and predict which accounts engage in fraud (or are more likely to engage in a future fraudulent transaction).

If anyone has today’s paper, the article in on page B3.

dbldie55

Sounds pretty clear to me.

Bear

"TRUST NO ONE.........SUSPECT, EVERYONE"

BigDaddyz

That's exactly what I have been complaining to ebay about when I send in fraud alerts. With the information they have at hand, they should be able to spot a huge amount of fraud without it going so far that they get recognized and reported by other users.


It should be very simple for them to do with programs what we do with simple observation.

Pentimento

From the article:

"We ran the algorithm over a large set of data from eBay, and we were able to locate some obvious cases of fraud that hadn't been spotted before."

I have seen forum members here team up and accomplish the same thing during a Sunday halftime. The fact remains that Ebay should be at the forefront of this work, not Carnegie Mellon grad students or coin collectors.

Very interesting link-- thanks!

mr1931S

tried to find article online and was unsuccessful...math nerds unite!

flaminio

Couldn't find the article online, but here's a link to the original paper. Sounds very complicated...

Millertime

Sounds like a NUMB3RS episode.

Millertime

Petescorner

Great! Now if anyone at eBay only cared about doing anything about it...

mr1931S

Indeed, "deep hole" data mines will reveal their "nuggets of truth" when subjected to sophisticated mathematical analysis...thanks, flaminio

LanLord

Two things come to mind:

1 - Ebay needs to get control of their world
B - It shouldn't be up to people that are losing money to prove they've been wronged, Ebay should have be held to the same controls that CEOs and CFOs are with the SOX requirements
4 - It's just sad that some students did this and Ebay couldn't

StampAlarm

ttt

Longacre

I tried to read the paper that Flamino linked. Man, I better stick to tax law. That paper was Greek to me and I had to stop reading.

ziggy29

<< <i>"We ran the algorithm over a large set of data from eBay, and we were able to locate some obvious cases of fraud that hadn't been spotted before." >>

If Category = "Coins" and Seller_Location = "China" then Fraud = "Yes"

sonofagunk

This is nothing new. I use to do it for a job. We use similar algorthms to make sure that our traders are not doing anything illegal. There are a TON of simple tests that you could do and catch ebay people (I do it all of the time and report them, but 75% of the time, nothing happens). If I had acccess to IP addresses of where bids were placed, I could catch a ton more.

They just do not want to catch people

Dentuck

There are three kinds of people on eBay: those who can count, and those who can't.

Longacre, I owe you an email on an unrelated topic. Could you PM me your address?

ziggy29

<< <i>This is nothing new. I use to do it for a job. We use similar algorthms to make sure that our traders are not doing anything illegal. There are a TON of simple tests that you could do and catch ebay people (I do it all of the time and report them, but 75% of the time, nothing happens). If I had acccess to IP addresses of where bids were placed, I could catch a ton more. >>

This reminds me of some of the discussions we had on heuristics when I was taking some CS graduate courses. This would have been an interesting case study to talk about in the lectures.

itsnotjustme

With a group size in the millions, and possible fraud networks of 5-10-15-20 accounts, I suspect it is hard to specify a group that doesn't deal with another group, as most e-bay accounts only deal with a very small percentage of the exsting e-bay accounts.

Regarding 3 types of people on e-bay...

There are 10 types of nerds in the world. Those that know binary and those that don't.