I have been working on detecting spammers on social networks nowadays. Social networks detect spam and spammers by using mostly users’ activities and spam reports. If you read spam and abuse policies (i.e. Twitter’s), they clearly say how an account flagged as spammer. But, in your own services, you do not have the power like theirs. Let’s see Twitter’s. The only usable conditions similar with Twitter are
- if you have followed and/or unfollowed large amounts of users in a short time period, particularly by automated means (aggressive following or follower churn)
my comment: we probably can’t measure because we can’t repeatedly update users’ information
- if you repeatedly create false or misleading content in an attempt to bring attention to an account, service or link
my comment: we can’t get old tweets and check links that user posted
- if you post misleading links (e.g. affiliate links, links to malware/click jacking pages, etc.)
my comment: we probably can’t decide the content if it’s misleading or not without using 3rd party services
As you can see, sadly they are not so usable for us. And since you can not monitor neither user activities nor users’ all posts including old ones you have to detect spam with what you have. So there are some strong challenges on it.
- Social spammer behaviors change too fast.
- A system that is capable of capturing most of the spams this month may fail to do so next month.
- Spammers get smarter and can create new, more organic accounts to avoid being detected.
- Once the spammers see that their fake accounts are caught by the system, and they come up with a new strategy to deceive the system.
- The spammers do not form a cluster themselves and they are well integrated into a larger social network.
- Each social network has it’s own spam culture. One’s solution usually does not fit into another’s.
As you see, it’s very hard to detect spam and spammers on social networks. The challenges are not simple things you can achieve with maths. The solution is a bit complicated and I will write about it later. Lastly my advice is that if you want to detect spammers on social networks, you should be a spammer first.
P.S. just kidding, don’t spam. Just do whatever you can do to know them very well and to think like them.