DCortesi . blog

OK, I figure it’s time to throw my hat into the ring.

I’ve posted in the past about Twitter spam and I run what I think to be a pretty fun website about Twitter Stats, but there seems to be a lot of conversation recently about Twitter and the noise ratio.

Obviously, people are trying to figure out how best to use Twitter given its recent surge in popularity and accompanying spaminess. Louis Gray made a blog post about his noise ratio and Stowe Boyd followed up with a post about the noise ratio and conversational index, but there’s one thing that seems to be common across both these posts:

There is a super-fantastic problem in that both posts discuss one, one ratio!

That’s right - one ratio to describe the entire activity of Twitterites. One ratio to rule them all, one ratio to find them, one ratio to bring them all and in the darkness bind them.

OK, perhaps these posts were intended to be their own personal way of determining a proper reputation structure on Twitter, but there is so much more data available to play with. Shall we? Yes, let’s take a look at all the numbers we have to play with:

• Friends
• Followers
• Favorites
• Updates
• Date joined Twitter
• Number of updates over time
• Number of updates in the past month vs. when the first joined Twitter
• % of updates that contain links
• % of updates that are replies
• Number of mentions of the word “awesome”

These are just a few of the numbers that Twitter provides and while the noise ratio is a nice statistics, it is most definitely not a holistic means of providing a method by which to rate the reputation of a Twitter user. And there never will be such a means. Myself and @wardspan had a conversation this evening where we discussed the top three things we use to determine if we’re going to follow somebody. I think we only shared one in common of our top 3 and we tend to be pretty similar-minded. But we use Twitter for different reasons.

And it is with this post that I call out for a reasonable reputation system across our many services. Twitter is one such example, but there have been others in the past (yes, those other social networks) that have dealt with the same reputational issue, not to mention spam.

And it’s not getting better. I signed up for FriendFeed today and created a profile of my real self’s online activity. The scary thing is…I could have created the same profile for anybody else and the question to ask yourself is would anybody have known any better? In addition, in their case - does it even matter? Or are they redirecting their trust to the other systems they are using to generate their content.

Just imagine, if we could create a reliable reputation system across the services that we use to provide us with better and more interesting, targeted content on a daily basis. If only…

A brief update - top 3 things that can be done to help users weed out spam:

1. Make the block functionality more accessible - did you find it underneath the “Following” legend?
2. Provide basic stats about a user in the notification email - location, bio and some ratio information
3. Use backend monitoring/analysis to `killall -9` spammer accounts (block ratio, usage trends indicative of automation, etc)

As with any social network, spammers appear to take advantage of the collective masses that are gathered and interacting with each other. This is no different on Twitter, where numerous people have complained recently about massive follows from spam accounts. These accounts typically take the form of a high following:friend ratio and a low number of updates. There is even a site devoted to Twitter spam, twitterspam.com. There’s quite a bit of other information we can examine, but let’s tackle this in order of the two main types of spam I’ve come across.

The first is embodied in the @castlebaths account. Statistics that indicate this as a possible spam account:

• 20% of links in the first 20 updates are the same as the bio link
• There are zero replies in the account (note: not unlike a new Twitter user)
• There’s an average of 1.15 updates/follower
• The users “Friends” account for 95% of the aggregate friends and followers

Now this account may very well be legitimate, but I doubt many people want to follow somebody on Twitter that is simply hawking a product and not contributing much beyond that. Taking these values and creating an aggregate score would probably score pretty high on the spam card.

Let’s take a look at another account, @kendra2. This account is a little bit more difficult to identify as spam through the numbers:

• 5% of the urls in the first 20 updates are the same as the bio link (that’s one url for those not counting)
• This account has actually replied to people
• There are only 14 updates, but
• The users “Friends” account for 95% of the aggregate friends and followers

This is an interesting account since it seems to be an actual person trying to interact, but the bio link is actually the telltale sign here - videochatonline is a webcam site and @kendra2 is obviously trying to bring traffic to that site. The numbers do not clearly mark this as spam, but the last two statistics seem to indicate this account has been created solely for the purpose driving traffic outside of Twitter. Other signs are the “pretty girl” avatar, bio link to a commercial site and potentially similar profiles.

As a Twitter user, what other statistics can I use to identify spam that Twitter (or somebody else…) might be able to provide?

• # of my friends that _also_ follow the account
• # of accounts without autofollow that are following the account
• # of inactive accounts being followed by the new user
• Are consecutive accounts being followed?

There’s also a number of back end statistics that can be utilized by Twitter such as unique IP addresses in use across large numbers accounts, clickstream rates and patterns and other similarities across multiple accounts. Reporting spam isn’t always useful, but observing the (generally predictable) behavior of spammers and the interaction of the users with those accounts is a step forward.

Is spam an easy problem? Obviously not or we wouldn’t have blog, email, trackback, comment and postal spam. Will there be false positives? Sure. However the numbers above can help in both the automatic identification of spam accounts and providing users with enough topical information to make smart decisions to help alleviate their frustration as well. Furnishing an easy means by which to report/block spam is also a necessary evil. Twitter has hummed along relatively under the spam radar until now, but it seems it has to accept that spammers will try to take advantage of its users. Giving users the power to identify and avoid spam through the use of statistics will hopefully make Twitter a fruitless source of successful spam.