I just came back from the monthly NY Tech Meetup, whose theme this evening was “Built on Twitter“. While the meeting was well organized (a testament to Nate Westheimer, who received the torch from Meetup CEO Scott Heiferman, I had mixed feelings about the demos. Everyone is capitalizing on Twitter’s buzz, but so few people seem to be creating anything valuable on top of it.
But, by luck, Daniel Lemire sent me a link to Sylvie Noël’s post about a paper by HP Labs on “Twitter: Social Networks that Matter: Twitter under the microscope” by Bernardo A. Huberman, Daniel M. Romero and Fang Wu. She also pointed to an executive summary by Forrester analyst Jeremiah Owyang.
The paper is insightful. The authors practically had me at hello–this is the paper’s third paragraph:
While the standard definition of a social network embodies the notion of all the people with whom one shares a social relationship, in reality people interact with very few of those “listed” as part of their network. One important reason behind this fact is that attention is the scarce resource in the age of the web. Users faced with many daily tasks and large number of social links default to interacting with those few that matter and that reciprocate their attention. For example, a recent study of Facebook showed that users only poke and message a small number of people while they have a large number of declared friends. And a casual search through recent calls made through any mobile phone usually reveals that a small percentage of the contacts stored in the phone are frequently contacted by the user.
They then define a user’s “friend” as a person to whom that user has specifically directed at least two posts and show that the a user’s number of friends is a better predictor of the user’s activity (number of posts) than the user’s number of followers. Having thus validated the number of friends as a more important input variable than the number of followers, they explore the friend graph, which turns out to be much sparser than the follower graph.
Many people, including scholars, advertisers and political activists, see online social networks as an opportunity to study the propagation of ideas, the formation of social bonds and viral marketing, among others. This view should be tempered by our findings that a link between any two people does not necessarily imply an interaction between them. As we showed in the case of Twitter, most of the links declared within Twitter were meaningless from an interaction point of view. Thus the need to find the hidden social network; the one that matters when trying to rely on word of mouth to spread an idea, a belief, or a trend.
I urge you to read the whole paper, as my abbreviated version hardly does it justice. And then, if you’re practically minded, think about ways to build applications on Twitter than leverage this real social network that is hidden in plain sight.
I further suspect that the authors result generalize beyond Twitter to other social networks where the cost of connecting is far lower than the cost of actually investing in the connection. It doesn’t seem hard to identify the hidden social network, and by doing so we can unlock its value.
Of course, Twitter has the virtue that its network is mostly available to the public, not hidden behind a walled garden like LinkedIn or Facebook. As a result, I expect that Twitter will drive both research and innovation in the social network space, at least in the near term.
23 replies on “The Real Twitter”
It’s a little surprising and disappointing to me that this paper should be such an eye-opener to people. I have always conceptualized social network (services) in this way. There’s who we know and who we do things with, and those aren’t necessarily the same sets.
The latter set may also change depending on the context. I talk to people about social network research, and to others about running, and to others about working on bicycles. While there is some overlap, to me these are three conceptually different graphs.
As for doing something useful with Twitter (and other social network-y apps), I’d also be excited to see something that lets people get things done in a directed, but distributed and self-organizing manner. The question I’d like to see answered is, how do we use social networks to do a particular thing? It may be that we can’t, that such systems are good at what they’re good at but that you can’t really choose, or there might be a way to get people to cooperate in order to achieve some goal. It probably depends on the task as much as the underlying network and incentive structure.
I’m not sure that the split in the paper is quite the one you’re making. I see the problem with online social networks as being that they don’t model the scarcity resource of our attention. I’d hoped Twitter would be different, but I was sorely naive and have since blogged about my disillusionment. What I like about this paper is that it wipes away the noise by identifying behavior that reflects investment of attention and thus brings back the scarcity constraint.
This isn’t quite “who we know” vs. “who we do things with”. Unless knowing someone means nothing more than knowing there Twitter id.
The last bit. I guess what I mean by “who we know” is just these big friend lists people keep around. In other words, we have these lists, and divide our attention amongst the various people in these lists. Some get more, some get less. In the end, there are only twenty-four hours in a day. I do believe that directed interaction can tell us a lot more about what relationships (edges) are durable than any self declared friend list.
What do you mean when you say online social networks don’t model the scarcity resource of our attention? Are you talking about the services they offer, or the sorts of data we tend to extract from them? I think you mean the former, but I’m not sure.
Maybe we’re just talking past each other. I agree with you that actual interaction is what counts. I perhaps go further than you in that I think a self-declared friend list is garbage if it does not correlate to actual interaction. After all, I could declare the entire phone book to be my friends.
When I say that online social networks don’t model the scarcity resource of our attention, what I mean is that they don’t reflect the fact that each person has a finite amount of attention to partition among his or her connections.
In the offline world, we normally don’t consider people “friends” or “colleagues” unless we allocate some minimum amount of our time to them. I don’t see why that should be different in the online world.
There’s certainly room for a finer-grained representation of our relationships. But I’d be content to start by culling the noise of meaningless “Barney” relationships where nothing of significance transpires.
Yeah, I don’t disagree with anything you’re saying, except that I wouldn’t necessarily put the onus of cleaning up friend lists on the services themselves.
I think I’m a little more lenient on friend lists, but that might just be because I put so little stock in them to begin with. As a researcher, I certainly would not rely on self-reported friend lists to induce a meaningful social graph. As a developer, though, the friend list seems like a useful mechanism (or, at least, a first cut) for putting some bounds on what information to present a user.
I think we’re in agreement that truly innovative social applications will go beyond the friend list and use behavioral patterns to organize information and interaction.
Yes, we agree on the problem and what’s generally needed to solve it, and I’m not particular about who does the solving.
That said, I’d think that online social networks would step up out of self-interest. They could provide far more value if they identified the more meaningful network hidden in the noise that they currently expose to their users. Users may be complicit–perhaps even primarily responsible–in creating the current state of affairs. But perhaps users are stuck in some tragedy of the commons where they feel incented to keep adding connections and thus devalue connections in the process–a sort of collectively inflicted inflation. The online social networks might not be obligated to solve the problem, but they may be best positioned to do so by changing the incentive structure.
I guess I don’t really see the problem with friend lists, even if they consist of largely bogus friends. If the real value of social network *data* is in the actual interactions between individuals, does it really matter to our apps and algorithms what people do with their friend lists?
Now, if we are talking about making the friend lists themselves a useful tool… then I could see a strong case for building incentive structures to only include active or “real” connections. On the other hand, wouldn’t it be better to just let users do what they will and mine that behavior data to construct a sort of filtered friend list for those who care?
Perhaps I am, as Curt Monash suggested, too much of an idealist. Why should an application encourage people to build lists of bogus friends? If they are going to leverage a real social network buried in the noise, why not expose the network so that people see the value in the real network and the lack of value in the bogus one?
But I think we’ve converged to a pretty similar stance. I would like to see social networking tools that incent social behavior and disincent anti-social behavior. I think that people creating large bogus friend lists by spamming the world is anti-social, but I suppose that is just my personal opinion.
Great conversation. Let me play devil’s advocate for a moment.
There is one benefit in following people even if you don’t interact with them: the ability to learn from them. For instance, I follow a bunch of people on twitter (e.g., Fred Wilson, Tim O’Reilly) as I find there thoughts help shape my view of the world. However, there is zero conversation going on as I never @ them.
This is mixing metaphors a bit, but I’ve read hundreds of posts on this blog but this is the first time I’ve posted. I’m not part of your ‘social network’, but you are definitely a part of mine.
I think the more subtle and challenging problem is to create social networks where the ‘expert’ or ‘celebrity’ can emerge while those who don’t want to be ‘experts’ or ‘celebrities’ (but may want to follow them) can continue to have meaningful relationships only with their close friends.
In other words, the hidden social network is hugely valuable, but so is the ‘celebrity’. The problem is that the middle ground is worthless – and how do you build services that let people avoid the middle ground.
Lindsay, that’s a fair point. I agree that there’s value in following people you don’t interact with. I certainly don’t restrict my own attention to those people who reciprocate.
But what bothers me about Twitter is that a lot of “following” seems empty, as I ranted in my earlier post about Twitter being an attention Ponzi scheme. Following should imply investment of actual attention, not just donation of empty status.
As for your last point, I’m hopeful that, once people get over the current desire to “get famous quick”, it will be easier to identify who are the experts / celebrities they care about. And it shouldn’t be that hard to know who your close friends are. We just have to get past this phase where we’re being drowned in noise.
Daniel – I think this is where we differ a bit. From a user perspective, I don’t think the rabid friending is too much of an issue. Those who really use such services will quickly realize that they need to manage their information, and real durable links will rise to the top.
Less important relationships (and people) should fall by the wayside as it becomes clear who participates/contributes, and who is just there to watch.
I think Lindsay is bringing up a really important point. These services are first and foremost user-centric. Their main purpose is to serve some user need, and that need is not necessarily to keep a record of a social graph. Like I said before, any real durable relationships will become evident through action, rather than by declaration.
Perhaps our difference is that you trust users to make the best choices for themselves, while I feel that system design influences the social norms in a way that, in practice, affects utility for users.
In particular, I believe that more people would use Twitter effectively if it offered a concept that mapped to a non-zero investment of attention. Maybe everyone will eventually figure it out. But it’s just as likely that the system will implode under its own weight first.
Following someone on Twitter has an opportunity cost in the default web client; if you follow someone, you see their tweets, and that might crowd your real friends tweets off the page.
Other Twitter clients (e.g. Tweetdeck) deal with that by doing client-side groups that let you slice attention into slices without letting the global world know.
There is behavior where people look to do attention cascades – they follow a lot of people, and behave to create new followers by begging for followers, and then occasionally go on a campaign to create some particular traffic to a particular site. There’s a fine line between marketing and spam…
Indeed, many Twitter clients allow you to “fake follow” people–that is, to give them the status benefit of a follower without incurring the opportunity cost of actually paying any attention to them. FriendFeed even goes as far as to support for this behavior is as a feature.
And the cascade-inciting behavior you cite is precisely what makes me call it an attention Ponzi scheme. Or perhaps a pump & dump metaphor is more appropriate.
My brain cells hurt as I try to sort through why I want to redefine what you see as a problem, Daniel. This is a wonderfully high-level discussion & I’ve been out of grad school for a while…. Here’s a first cut at my thoughts.
You seem to want users to do something specific–extract “real” social network value, rather than just doing spammy following.
Some will, some won’t.
The decision of someone to follow 2,000 people, hoping that some will follow back and click the junk in their autofollow DM, in no way affects MY utility in doing just what Lindsey describes–learning from someone who may or may not return the attention I’ve chosen to give. (And I definitely recognize the cost is not zero.) I’m not in the Ponzi scheme of the pump & dump folks because I don’t follow them.
Nor does the spammy stuff affect my ability to do what you suggest, which is to build a real social network with real interactions, and real value that I receive for my investment of attention.
Twitter does impose the 2,000 follow limitation; people have to attract 1,600 followers in order to exceed that. So you have to provide something, to someone who has chosen to define it as value, in order to expand your connections past a certain threshold.
I don’t think there’s one common, operationalizable definition of “value” you can use to put this into an equation.
This is a tool. Some people will build really crappy structures, some will build beautiful sustainable LEED-certified structures. Same tool, different results based on skill level of the craftsperson.
Nor do I think you can come up with THE attention management system to restructure any interaction space whether it’s Twitter or email. People find a way to be gullible in any communications space. Watched late-night TV ads for male enhancement supplements lately? Do they make Discovery Channel’s Planet Earth less valuable?
Another challenge: Like John, I’m following multiple interests and extracting different types of value, some of which is invisible to you in looking at the network in my primary personal Twitter account. I manage 4 other accounts, 3 for volunteer groups and one for work. You’ll never know where I see value and invest attention because it’s not all in one account.
Nor is it all in one social network, for that matter. If I see a blog post on Twitter and email it to my coworkers, that’s invisible (even to Google, despite @ikepigott’s scary piece).
OK, stopping at last. This is fascinating, thought-provoking, and sort of like yoga for my neurons.
Barb, thanks for sharing your thoughts! As you said on Twitter, I really am an idealist. And maybe I’m going to far in imposing a value system on users who are content to use Twitter for spammy following.
But I don’t think users are so much content as confused. No one sets out to participate in a Ponzi scheme. Rather, they think they’re going to get rich quick and their greed overcomes their rationality. And this is how I interpret the actions of people who invest time in accumulating followers just for the sake of raising their follower counts.
And I disagree that other people’s decisions on whom to follow don’t affect you. At least they affect me.
For example, I look at someone following me as a signal that they’re investing attention in my ideas, much as a person I’m talking with face to face is listening to me. But I know that a person can no more follow 1,000 people than listen to than many speaking in a cacophony.
Also, perhaps thinking more of a system like LinkedIn than Twitter, I’d love to be able to explore the social network to discover people with whom I have things in common. That’s much easier to do when the links represent real connections. Spammy following degrades the network as a whole.
I’m idealist but not a totalitarian: I don’t claim that users shouldn’t have the right to follow whom the please. I just want the system to maintain a representation with some verisimilitude to meaningful social connection.
I think the analysis of the network should start with smaller inner-networks within the larger network as these tend to behave more like human networks do now.
In these graphs the people we follow and hardly ever converse with because they are interesting will show as fairly consistent outliers and carry that value through the tweet rank.
Active members (multi-way) communication can then be modeled in each sub-graph to show where twitters rank for each individual twitterer (you, me, etc).
Finally these findings can be applied to intersecting and encompassing graphs to potentially rank people similarly even if they have an inordinate amount of followers.
Came to this post a little late as I read this tweet (http://bit.ly/Rx4C) but area of interest & some small scale work so thought I’d comment.
P.S. When doing the analysis let us not forget the people who comment on lots of others work but rarely start a 2-way interaction. I think I captured that above but just thought I would mini-clarify.
I actually think it makes sense to unify retweeting, commenting, and replying in a unified notion of propagating the signal of the message that drives these actions. At least that’s what I had in mind when I was proposing my “TunkRank” measure:
I don’t know enough about the properties of the de-noised Twitter graph (i.e., the sparse real graph of interactions underlying the dense one naively derived from follower links) to understand how best to analyze it. That’s why I started by trying to get a simple but plausible measure of influence. But I could certainly see taking it further to analyze similarity, interactions over time, etc.
I promise I’ll look into this if someone implements TunkRank!
[…] other day, I talked about the “real” Twitter: the sparse subgraph of meaningful social relationships buried in the far denser follower graph. […]
[…] on Twitter works well for me, but that’s probably because I have a substantial number of real, knowledgeable followers (the TunkRank is strong with this […]
I suspect you’re overlooking the value of weak ties. I’m sure I only interact regularly with a very small percentage of the people I’m listed as connected to on Facebook, but the specific people involved shift around a bit, and I occasionally make serendipitous connections with people two hops away from me as we both, say, comment in a public conversation based off of a mutual friend’s status message.
So I think the larger “phony” network could possibly be think of a latent network, somewhat available (in part) for activation.
Xian, I recognize the value of weak ties–but here, we’re talking about *no* ties–people who never interact with each other. Regular interaction is understandably sparse, but I think it’s shocking that having any interaction at all is sparse.
I grant the theoretical possibility that there’s a latent network, but I think the simpler explanation is that it’s phony. I’m not sure what would be needed to prove one hypothesis or the other.