Algorithms for flagging scam emails sent through a website

So I have an ASP.NET MVC 3 website that allows people to advertise certain items they have for sale. The general public can come to the site and browse the ads. On the detail page of each ad there is a form to the right that allows you to respond to an ad without having to create an account.

I've recently been getting complaints from some users that they're receiving ad responses such as "I am interested in the item, please I will send to you a cashiers check in the amount of $2,000 USD over your asking price..." etc. You've all seen the scam.

I want to be able to leave the ad response form available to people to use without having to create an account (that's just a roadblock) but at the same time implement whatever security measures I can to crack down on this sort of stuff.

So my question is: what suggestions do people have for something like this? I've already implemented CAPTCHA and throttled the number of ad responses an IP address can submit within certain time periods. Based on the types of ads, I don't s开发者_C百科ee how someone legitimately using the site would exceed the limits, but it would cause spammers to have to sit and wait before being able to send another scam so their time is better spent elsewhere.

What other "behind the scenes" algorithms have people used? I'm thinking of things like comparing responses from the same person and rejecting them if they're a certain percentage match (so you can't copy/paste the exact same response time after time).

Are there established ways of checking the content of the email against some sort of "scam dictionary" to see if the content of the email might constitute a scam? If I could detect that, I could flag the email for review before releasing it to the customer.

These are just some of the ideas I had but I was curious to know what others have implemented. I'm working in C#, so if anyone wants to provide samples of what they've done, that might prove useful as well.

BTW - not sure how to convert this to a wiki, but it seems like it might be a better fit for the type of question I'm asking.

You could allow your users to flag mails as spam, then use your messages as a training corpus for a learning system - for example, by using Google's Prediction API.

There are databases and services available that help identify spam messages, see for example Akismet:

http://akismet.com/