Popularity Is Irrelevant, Says New Measure of Influence on Twitter

A unique analysis of influence on Twitter shows that an obscure Indonesian filmmaker is just as good at propagating memes as @Google.

Christopher Mimsarchive page

August 5, 2010

The same team of HP researchers who proved that buzz on Twitter is an accurate predictor of box office sales for new movies has calculated a new way to measure influence on the micro-blogging social network.

Their algorithm turns out to be far better at predicting how far a link will travel than counting followers, and is even better than the PageRank algorithm that powers the search results delivered by Google.

It’s called IP-Influence, and its predictive power reveals two facts important to anyone who wants to spread their message on Twitter:

1. The overwhelming majority of people on Twitter are passive - that is, they rarely if ever retweet anything.

2. The best predictor of how far a tweet or link will travel on Twitter is how much power its originator has to motivate the most passive of his or her followers to retweet it.

A user’s IP-Influence does not correlate well with the number of followers they have, which explains why, among Twitter’s most influential handles, Indonesian filmmaker Joko Anwar stands shoulder-to-shoulder with Google. Here, according to the HP analysis, are Twitter’s most influential accounts, as measured by their ability to get their links retweeted and subsequently clicked on:

@mashable, @jokoanwar, @google, @aplusk, @syfy, @smashingmag, @michellemalkin,@theonion, @rww, @breakingnews

In the extreme case, SyFy has only 40,000 followers on Twitter, or less 1% of the followers of any of the most popular users on twitter, which just goes to show that Nerds > Justin Bieber any damn day of the week.

Future applications of this algorithm could allow savvy twitter watchers to accurately predict what topics and links are going to go viral on Twitter before they actually do. From the paper:

Content ranking. The predictive power of IP-influence can be used for content filtering and ranking in order to reveal content that is most likely to receive attention based on which users mentioned that content early on. Similarly, as in the case of users, this can be computed on a per-topic or per-user-group basis.

h/t Chris Messina

Follow Mims on Twitter or contact him via email.

Keep Reading

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.