Babbage | Popularity does not equal influence on Twitter

Seeking influence

Popularity does not equal influence on Twitter

By L.S.

PEOPLE who are popular have a lot of influence. Or so Babbage thought. But things are not that simple, as research by the Social Computing Lab at Hewlett-Packard's research arm shows.

Bernardo Huberman, the lab's director, and his colleagues (Daniel Romerso, Wojciech Galuba and Sitaram Asur)analysed 22m messages on Twitter, the micro-blogging service, to find out, among other things, how popularity and influence correlate. Measuring popularity is straightforward: it rises with the number of those who have signed up to follow a person's or organisation's messages. In contrast, determining influence is more tricky. The researchers hypothesised that users of Twitter are the more influential the more they manage to overcome their followers' "passivity", meaning their tendency not to to pass on messages (only one in 318 messages containing an internet address is "retweeted" the researchers found).

Based on that assumption they developed an algorithm that calculates a user's influence. The most surprising result is that the correlation between popularity and influence is weaker than one might expect. Some media, for instance, have lots of followers, but don't do so well when it comes to having their messages re-tweeted. People Magazine and The Onion, a satirical website, had about the same number of followers in September 2009 when the dataset was obtained (2,1m versus 2,3m). But the former ranked 565th in terms of influence and the later 92th. And The Economist, which had only 311,109 followers, ranked a respectable 902 in influence. Below is the list of the 30 most influential media on Twitter (the entire list can be found here):

Number@nameNameFollowersInfluence Rank
1@mashablePete Cashmore203784059
2@cnnbrkCNN Breaking News322447571
3@big_pictureThe Big Picture2366692
4@theonionThe Onion2289939116
5@timeTIME.com2111832143
6@breakingnewsBreaking News1795976147
7@bbcbreakingBBC Breaking News509756168
8@espnESPN572577187
9@harvardbizHarvard Business Rev219039227
10@gizmodoGizmodo111025237
11@techcrunchTechCrunch1402254319
12@wiredWired547187322
13@wsjWall Street Journal366133358
14@smashingmagSmashing Magazine224333360
15@pitchforkmediaPitchfork1494896384
16@rollingstoneRolling Stone133999436
17@whitehouseThe White House1794544448
18@cnnCNN1196719473
19@tweetmemeTweetMeme52386515
20@peoplemagPeople magazine2099081565
21@natgeosocietyNational Geographic274851603
22@nytimesThe New York Times2502914705
23@lifehackerLifehacker62302708
24@foxnewsFox News260081710
25@waitwaitwaitwait32895824
26@newsweekNewsweek1250884844
27@huffingtonpostHuffington Post632555849
28@newscientistNew Scientist144355852
29@mental_flossMental Floss68975874
30@theeconomistThe Economist311109

902

With politicians, the disconnect is even more pronounced, at least in America. If Mr Huberman and his colleagues got their numbers right, the most influencial American politician on Twitter is not John McCain (who has 1,7m followers), but Nancy Pelosi (15,964). Yet overall the results are rather good news for the Republicans: 70 of the 100 most influential members of Congress are from the GOP.

Experts will surely debate at length whether the algorithm designed by HP's Social Computing Lab is a good one for measuring influence. Others will ask: So what? But the exercise is far from futile. In a world that is increasingly swamped by tweets, posts, likes and so on, it comes in handy to know who is listened to. In fact, it is crucial if firms, governments and individuals want to get their message through on social networks, Mr Huberman and his colleagues argue in the paper that they have written an their research (which they have posted here).

Other researchers will certainly devise alternative ways of measuring influence. Given enough money, they will have an easier time than their colleagues at HP. To get the raw data, HP had to query Twitter's search function for 300 hours. But on November 17th it emerged that the service will sell access to its messages via a partner, Gnip. The firm will reportedly provide half of all tweets for $360,000 per year, or 5% for $60,000. Even the latter offer provides more than enough data to find the most influential needles in the digital haystack: currently about 5m messages per day.

Discover more

And it’s goodnight from us

Why 10, not 9, is better than 8

For Microsoft, Windows 10 is both the end of the line and a new beginning


Future, imperfect and tense

Deadlines in the future are more likely to be met if they are linked to the mind's slippery notions of the present