Like any communication platform, Twitter faces problems with spam.
For Twitter these issues lie in 3 main areas:
Large volumes of spam Tweets (particularly @mentions)
Fake accounts with millions of followers
Direct Message (DM) spam from fake accounts
Twitter have done a reasonable job at squashing spam Tweets.While it’s really hard to detect spam in 140 characters, all links included in Tweets are passed through Twitter’s t.co shortener. This allows them to retroactively mark links as phishing or spam which has really helped to limit the ability of spammers to get much out of large scale spam campaigns. As soon as a domain or URL is detected as compromised, all sent Tweets become useless.
However, Twitter have made little headway into reducing the number of fake Twitter accounts that exist. There is a large black market where you can buy a million fake followers for $2500 USD or less. These accounts are a problem not only because they falsify account influence indicators, but because they enable DM spam. These accounts look legitimate but when you follow them back, they now have access to send you private messages containing spam.
The problem of automated DMs and DM spam has become so large that most active Twitter accounts find the DM system unusable. They will never check their DM inboxes and turn off all notifications that a DM has been received.
Fake accounts build their followers through 2 mechanisms:
Registering a large number of accounts and having them follow each other
Following a large number of legitimate accounts so that some of them will follow them back. This has the added bonus of providing a path to DM spam.
Twitter have recently taken a stab at this problem with the changes to their automation and bulk following policy.
At a high level what they’re doing appears to make sense. If fake accounts are built with automation, then stopping all apps from doing any kind of automation should kill off this ecosystem right? Not so simple.
The problem is that spam accounts don’t use legitimate Twitter apps like ManageFlitter to build their accounts. Twitter has a range of restrictions already in place around automation that make it quite inefficient to use build accounts without a lot of manual input and selection. We monitor usage of our system closely in case we need to suspend accounts, but we rarely need to. We know that fake accounts are not using our system. There are a range of black-hat Twitter tools that side-step the API and act like web browsers, faking automated requests to Twitter in order to build huge numbers of followers.
The solution that Twitter has come up with means reaching into third party applications a restricting how parts of the UI look and work. This is a losing game that only inconveniences legitimate usage of Twitter.
The only way to solve the spam problem is to bring stronger limits to bear in the way social graph changes can be made on Twitter. If spam accounts are not using the API only solutions that span every possible usage of Twitter will work. This means hard limits inside Twitter on how many people an account can follow each day, or better yet changes to the way the existing limits work to better stamp out aggressive social graph activity.
Twitter has an existing limit that stop people from following more than 2,000 people unless a similar number of people are following them.
I propose that a solution to the fake account problem would be to lower this limit to 1,000 and to stop unfollows from reducing the limit for a period of time, say 7 days.
This process would look something like this:
@fake_account follows 1,000 people and hits the following limit
12 hours later @fake_account unfollows the 900 people who didn’t follow them back
Normally @fake_account would then be able to follow another 1,000 people straight away, however under the new system accounts would have to wait another 7 days before their unfollows were removed from their follow limit and they could begin following again.
This would severely limit the ability of accounts to follow huge numbers in bulk and repeatedly perform aggressive graph activity. It would encourage legitimate accounts trying to build a following organically to follow targeted people genuinely interested in their account and likely to follow them back. Verified accounts could be excluded from these limits.
Additionally we would like to see Twitter introduce better algorithms for detecting spam accounts.ManageFlitter has built a spam score which lets us detect how likely it is that an account is fake. Just by measuring a few simple indicators, and plugging them into a baysian probability algorithm, we’ve been able to detect spam accounts with a very high accuracy. If Twitter invested more time into building a similar tool, they could fight social graph spam in this way and suspend offending accounts without penalising legitimate users.
With ManageFlitter, we have built a set of tools to help brands and business build their social graph organically. We want to find a solution to the Twitter spam problem that still gives freedom to the way the businesses and brands are engaging and building their audiences. Twitter’s most recent change does nothing to address the real problems with spam. They’re simply inconveniencing their power users and making it harder for legitimate users to work efficiently on their platform. All of this effort from and yet the spammers will still be winning.
CTO & Co-Founder