- Dang, I thought I was following some of those people already
- Um, I recently unfollowed some of those folks
- I don’t engage much with some of the referral sources
- There are a lot of celebrities on this list
- These are real people, not feeds
Let’s look at these impressions individually and I’ll share some suggestions that will help Twitter’s suggestion algorithm.
1) Already following these people
No really, I thought I was already following some of the people that showed up. Is it possible people have been getting “disconnected” over the past few days? I saw the same reaction from others in my Twitter timeline. If Twitter “caught” the oversight then they did something right.
2) I recently unfollowed some of the suggested users
It happens. We unfollow people for one reason or another. Maybe they don’t engage with you, they say something that irks you, or you’re tired of following a particular celebrity. Many of the people who I had already followed and unfollowed were celebrities. I’ve already been to the show, I don’t need to go back.
Algorithm Improvement: De-prioritize accounts that one has already followed and unfollowed and especially those that have been blocked.
3) Don’t engage much with “Followed By” followers
Each Suggested User has several bits of information associated with her account – Name, Twitter handle, location, bio, and Followed By. All suggested users have two or more accounts in the “Followed By” field and I assume these are my main “connections” to the suggested user. But many of the people that appeared in the “Followed By” field where people who I don’t engage with on a consistent basis.
Algorithm Improvement: Follow Klout’s lead and make suggestions based on key connectors. Klout analyzes a user’s statistical influence and also lists users that are influential. These influencers should be the top “connectors” used in the Suggested User formula.
4) Lots of celebrities
This is where the asymmetrical nature of Twitter is a drawback to making equitable connections on the social graph. By asymmetrical, I mean that one can follow an account but the account doesn’t need to follow back. In contrast, Facebook requires people to follow each other to be “Friends” so that relationship is symmetrical. Extremely asymmetrical “nodes” in a network cause a warping in the fabric of the graph. If a celebrity (that doesn’t follow me or only a few others) suggests another celebrity that they follow then that’s a low probability match.
Algorithm Improvement:Turn down the volume on celebrity suggestions. These accounts will be easy to spot by the ratio of followers to people being followed.
5) These are real people
Every person on the Who to Follow list appear to be engaged Twitter users, not just piped in “feeds” from other social networking tools. Twitter got this right.
I follow 7,740 people on Twitter and am followed by 8,157 accounts and have new followers every day. I don’t use auto-followback tools so I have to manually go through my email messages and “vet” followers. Unless the account is extremely relevant to my interests, I don’t follow feed accounts. (This also means I’m way behind in reviewing new followers.) By filtering out feed-only accounts, this tremendously increases the chances of my following them. Good job Twitter.
Twitter still has some work to do on making the Suggestions more relevant. I made suggestions but also gave Twitter kudos on some things it is doing right. This is much better than the old Suggested Users celebfest hyper-monetization list it had before.
What do you think? Did you see any other obvious algorithm improvements? What was your overall impression of the Suggestions For You feature?
TechCrunch – Twitter’s Social Graph Is About To Get Pumped Up. “Who To Follow” Is Social Steroids