A group of Microsoft researchers has used supervised machine learning to try and improve detection of fraudulent user accounts. With Skype as their test platform, the group says it was able to achieve 68 per cent successful detection of fake accounts within four months of activity, while keeping false positives down to 5 per …
68% detection after ONLY four months?
Oh that will make a difference won't it, but I am a little confused, is that 68% of all accounts or 68% of what they looked at.
And if you account happens to be one of the 5% false positives?
Neither, the algorithm they built was able to detect 68% of the known-spam accounts and incorrectly identified 5% of the known-good accounts as spam.
The algorithm wasn't running for 4 months either, the data they were using was on accounts that have been active for 4 months without being flagged but were determined to be spam accounts.
...... have access to the usernames I would think that 68% detection on a blind study using the methods they did was quite good.
With the usernames available I would expect that to be in the very high 90's as at least once a week I get an invite or message from someone with enuslcenvfo or ggggggggggggv as their name.
Sod's law says there will be someone on here with that name.
> ggggggggggggv as their name.
Oh, sorry, I must have mistaken you for someone else.
68% maybe ok if it uses a methodology that has nearly no false positives but combined with a 1 in 20 false positive is pretty poor. You could probably get similar stats by assuming anyone who sends a URL is a spammer
I just had someone sign up for a skype account using my email address.
Turns out, they don't require email verification to set up a new account associated with an email address.
Might be a prudent first step.
Trouble is that if, say, only 1% of accounts are actually fake (it's probably much less) then their results will be swamped by false positives.
fscked by SHA-1 collision? Not so fast, says Linus Torvalds