Influence, network effects, and utility
Reputation Online yesterday looked at the “difficulties of algorithmic authority” in an article featuring quotes from Azeem Azhar, founder of Peer Index. As he did when I interviewed him for our “Identifying and targeting social media influencers” webinar, Azhar urged caution around the very idea of “influence”:
“What we measure is how people respond to each other [online] over time… Using a network structure, can you interpret who is more important to people, which sites are more important to people? Yes you can… We can tell you how authoritative, how much of an opinion-leader a person is in a given field.”
Importance? Authority? I’m not so sure these concepts are any less woolly than influence. They’re certainly susceptible to the kind of criticsms leveled by Vikki Chowney in a February NMA article:
… you still can’t assess influence based on online behaviour alone. First, you can lie online. You can steal people’s ideas, rehash them as your own, then create a perceived image of knowing what you’re talking about. Second, this doesn’t take offline activity into account. Similar to the notion of creating a persona online, you could be a complete tool in the real world…
Meeting Chowney in the comments, Azhar conceded that meatspace measurement is more problematic, but defended algorithmic assessments of online networks:
On a strict definition of influence, the academic research basically comes down to ‘you probably can’t identify the specific node which causes a message to spread’ [but y]ou can absolutely use social network analyses (and other analyses) to figure out which nodes in a network (for example people) sit across information flow..
Azhar is saying that you can identify the nodes through which particular content has spread and therefore deduce the specific nodes that will probably cause messages to spread in future. This analysis isn’t as facile as it sounds. For example, this recent paper from researchers at the University of Cambridge and Korea’s KAIST highlights the role of “power users” in the propagation of mainstream media content from individual journalists’ Twitter accounts.
It goes without saying I’m with Chowney in her wariness of, erm, “real-life tools” – although I’d say most of these are readily identifiable from their online activity – but as for the false identities and rehashed ideas, who really cares? If I find value in the information, I’ll happily heed an agency bot.
Conflating influence (or authority, or importance, or trust, or any of these oft-commingled concepts) with the movement of content through online networks might be more dangerous. It’s clearly reductive. But the approach’s value tends to be indistinguishable from its utility. I can think of one company whose influence ranking is seldom questioned no matter how blackbox it is.
Leave a Comment
Trackbacks & Pingbacks
[…] Well, quite, but close examination of 100,000 individuals is rarely achievable. Shortcut 101 for assessing the “influence” of armies of followers would be… the number of followers those followers have, and then the number of followers’ followers’ followers, and so on; a PageRank for Twitter (but not TweetRank). Add to that all the other channels through which individuals can potentially exert “influence”, and by necessity the analysis becomes less human, and more mathematical. As we’ve written here before, the proof of such an approach is very much in the pudding. […]