Whose Tweets Matter Most?
A startup monitors the spread of information on Twitter to identify users with the most influence.
It’s not the number of followers you have on Twitter that counts–it’s how you influence them. That’s the message from a new service called PeerIndex that analyzes the flow of information through Twitter. It offers a way to find people who are particularly authoritative in certain domains.
The users it turns up may have relatively few followers but can still wield a huge amount of influence– in particular subject areas.
The social Web has provided new ways to connect with people and discover information. But identifying important sources of information within all the chatter can be as time-consuming as ever.
The founder of PeerIndex, Azeem Azhar, who until recently was head of innovation at news agency Reuters, says his service could act as an intelligent Yellow Pages. For example, it could help a company find people to start work on a project in a specific area, or it could help PR firms spread news as widely and effectively as possible. “The way a company or person figures out who is an authority today is slow, bespoke, and expensive,” Azhar says. “We make it much easier.”
Twitter publishes statistics that show how many people a user is following, and how many users are following them. Those figures are used by third parties to create simple rankings of who has the most followers–topped by celebrities with millions of fans. Other rankings also use tallies of the number of times a person’s messages are “retweeted” to generate better measures of influence.
“Those indexes are very good at identifying the person who at a party would create a buzz around themselves,” says Azhar, but not experts who are quietly influential in their field. PeerIndex looks at the information contained in tweets, and how that information spreads, to find authority in specific subject areas. This provides a subtler measure of influence, he says.
For example, PeerIndex’s list of authorities on climate change is very different from that for Indian business, and these two lists are different from the ranking you would get by merely filtering the list of top tweeters by subject. “If you do it that way, a top user of Twitter with a lot of followers only needs to say something a couple of times to become an authority on a topic,” explains Azhar.
PeerIndex is based on a database of tweets harvested from Twitter–currently those written by around two million people. It connects all of these users in a network, or graph, according to how information in tweets is shared between them, and then tracks information flow on a particular topic by looking at how links, words, or phrases are picked up and reused by others. Mathematical features of that can reveal the people who introduce new information on a particular topic that spreads widely. These people are deemed authorities in that area. The current range of topics spans from urban renewal to venture capital; in the future users will be able to define custom topics.
“PeerIndex generates some nice results,” says Daniel Tunkelang, who has developed his own algorithm for ranking Twitter users that considers both a person’s follower count and the follower counts of their followers. His ranking is available through the site TunkRank. “The immediate application of such lists is suggesting followers for a given topic,” Tunkelang says. Authority measures could eventually be used to filter and organize the information people see on a social network, he says.
PeerIndex has a webpage for each person it has rated the authority of and invites people to connect theirs to their LinkedIn or Facebook profiles. It may be possible to apply similar authority measures to these social networks, says Azhar.
Tunkelang agrees that authority measures could add value to other networks. “I’d love to see LinkedIn quantify the expertise of the people in its network,” he says, adding that authority ranking is likely to become an important feature of question-answering communities like Quora.
An issue that any ranking system will need to cope with is spam–a growing problem on Twitter and other networks. Computer science professor Daniel Gayo-Avello at the University of Oviedo, in Spain, recently published a study showing how different Twitter rankings are skewed by spam accounts. “One of the lessons is that follower count is not a robust measure of authority, since it is easily manipulated,” says Tunkelang. His algorithm fared the best out of those tested by Gayo-Avello.
A content-centric approach like PeerIndex’s sidesteps the issue of inflated follower counts, but any ranking that becomes popular will likely attract attention from people hoping to game the system. “I think we have to expect an arms race the moment that any ranking algorithm becomes popular enough to be worth manipulating,” says Tunkelang.