Silk Road forums
Discussion => Newbie discussion => Topic started by: deebo on April 11, 2013, 07:04 pm
-
Hello SR,
I've been a guest lurking on these forums for the last couple of months but I figured it was finally time to join the conversation. I've completed 10 transactions totaling $5k over my 5 months on the road and I've had to deal with two scammers. (20%!!!) I'm a software engineer by trade who has a passion for finding innovative solutions to rudimentary problems and I'd like to propose a mathematically based approach to identifying scams for both buyers and vendors.
I believe that all user activity on SR could be very accurately modeled with what's known as a "Hidden Markov Model". Here's a quick crash course for HMMs:
You have a sequence of observations (user actions/feedback), which can be assumed to be governed by an unknown stochastic process(unknown intentions). The process is actually a Markov chain, which changes from state to state with unknown probabilities and each state emits an observation with a particular probability distribution (historical actions). Holding the number of states constant (good, evil), we can use our observations to approximate the starting probabilities of each state, and calculate the transition matrix as well as the emission probability matrix. It would then be possible to perform a "Monte Carlo" analysis or use Viterbi's algorithm to identify sketchy behavior as its occurring rather than after the fact.
These algorithms are extremely powerful. They are used to predict stock prices, find new gas reserves, predict gene expression, build better spacecraft and to analyze cryptographic communications.
Surely out of the massive SR community we have a handful of bright people that would be willing to come together to make something like this happen. Or at the very least create a tool for the administrators that would identify vendors or buyers who are most at risk and put them on a watch list. As the community grows the need for automation will too. It irks me to think of the hundreds of thousands that have been scammed from honest people on this site when there exists a tangible way to prevent the majority of it from ever happening.
Call me an idealist but lets employ the wonders of modern mathematics to make the road a safer place for everyone.
-Deebo
-
This is a very good idea, now it just needs to be implemented. 8)
-
This is an interesting and well presented topic.
I have to wonder though - if the algorithms are so powerful and accurate, wouldn't it be a far better use of resources to use them to uhm... say predict stock prices? Or better yet, Bitcoin prices!
Many really smart people have already spent lots of hours looking at the exact issues you mention - detecting scam artists, and predicting stock prices. They've made progress, but not much. No offense to the SR forums audience, but why do you thinks developers lurking on this newbie forum will overcome issues where others have failed? Have you even ever written a complex piece of software?
<b>And when is gets right down to it, detecting scams might best be performed by the best and most elusive algorithm of all - human intuition.</b>
:::: this post presented to you in the interest of circumventing the phenomenally ill-advised 50 post minimum for the big boy forums ::::
-
Now it just needs to be implemented.
I believe the greatest hurdle to innovation on SR is DPR's inability (for damn good reason) to trust those with the capability to implement fresh ideas. In order for my proposed system work, the algorithm would require access to every user's entire action history and deep integration into the server side software. How could DPR trust anyone with access to such sensitive information and source code even if only for development purposes? The same reasoning applies to any modification for the site, regardless of complexity.
I have to wonder though - if the algorithms are so powerful and accurate, wouldn't it be a far better use of resources to use them to uhm... say predict stock prices? Or better yet, Bitcoin prices!
I've been actively developing a program that uses a very closely related cousin to these algorithms to predict bitcoin exchange rates. I've made enough money this way to feel confident enough to quit my job. I hope to tweak it further to secure my retirement fund. I'm not the first person to do this. There are whole trading firms that use super computers running these algorithms to make the hedgefund trading decisions. Because they trade so quickly and work off of such small margins, automated trading has almost crashed the NASDAQ four times that I know of thus far.
It's not a "better use of resources", just another application in a new context.
-Deebo
-
This is an interesting concept, but I'm not sure how well this would work in the context of SR. I think a seller's honesty can best be inferred by reviewing feedback scores/comments (neglecting all FE comments), comparing prices to comparable SR products (avoiding the "too good to be true" listings), an incidence rate of disputes (and how they are resolved), etc.
I'm not sure that user actions really have the Markov property in reality. Generally speaking, a more experienced vendor will be less likely to scam, and a more experienced buyer will be less likely to scam. When a vendor scams, it's probably either selective scamming (tricky to detect / pinpoint), scamming from the beginning (which could be reasonably well detected by the aforementioned methods), or a SR vendor quitting and cashing out by scamming. Since lots of the scammers are new vendors, the memoryless property isn't really satisfied.
Anyhow, should you come up with anything, I'd be interested in reading more. I'm decently versed in statistics and computer science. I've also considered writing a bitcoin trader, but I'm not wealthy enough to justify risking some capital to make a bad-ass daytrader.