My name is danah boyd and I'm a Principal Researcher at Microsoft Research and the founder/president of Data & Society. Buzzwords in my world include: privacy, context, youth culture, social media, big data. I use this blog to express random thoughts about whatever I'm thinking.

Relevant links:

Archive

Attention Networks vs. Social Networks

(originally posted on centrality)

Network analysts often speak about (un)directed graphs. In essence, this refers to whether or not someone you know knows you. If reciprocity is required by the system, it’s an undirected graph. The vast majority of online social networking tools assume that users are modeling friendship and thus if you’re friends with someone, they better damn well be friends with you. As such, they use undirected graphs and you are required to confirm that they are indeed your friend.

Well, what about fandom? Orkut actually put the concept of fan into their system, but in order to be someone’s fan, you had to be their friend first. Baroo? I’ve noticed that Friendster introduced fans, although it is not consistent across the site; the system decides who is celebrity. I can be a fan of Pamela Anderson but i cannot be a fan of Michel Foucault or Henry Jenkins. While i can understand that the former is clearly a Fakester, the latter is actually a real academic with a Friendster Profile that i genuinely admire (far more than Ms. Anderson). Even on MySpace where bands have a separate section, i have to add them to my friends; i cannot simply be fans.

The world is not an undirected graph and very little about social life online is actually undirected. Many social relations are unequal; they are rooted in directional graphs – fandom, power, hierarchy. So why do we use undirected models?

Of course, there are many systems that have directed graphs. I can read blogs by bloggers who who don’t read me; blogrolls are directed. I can have friends on LiveJournal that do not reciprocate. I can subscribe to del.icio.us feeds of people that i admire without forcing them to do the same. I can make a Flickr user a contact simply so that i can watch their photos. I do all this because i know the world is not undirected.

Part of the problem is that we’ve built a model off of social networks instead of attention networks and there’s a very subtle difference between the two. Attention networks recognize power. They recognize that someone may actually have a good collection of references or be a good photographer and that someone else may want to pay attention to them even if their own collections are not worthy of reciprocation. Attention networks realize that the world is not an undirected graph.

There are many good reasons to use attention networks in systems instead of social networks. Do you really want to force people to get permission to subscribe to public material of someone else? Do you really want to put people through the awkwardness of having to approve someone that they don’t know simply because one person respects the other? Of course, the awkwardness of social networks does not disappear simply by having directed graphs. Reciprocity is still an issue whenever the networks are performative (visible as a statement of connection). This is most apparent in the blogging community where people feel insulted that they are not included on the blogroll of a blog that they read regularly. Thus, people feel the need to perform a relation of someone that they do not read simply for good social measure.

Attention networks are far more visible when people actually use the network for some purpose. Friendster networks are meant to be performative first and foremost. There’s minimal cost to having more friends. It may foul up your gallery searches but, really, does it make a difference if you see 4,325,935 people instead of 4,311,266? Attention networks like LiveJournal and Flickr combine the network with the subscription process. You want to keep your Friends page clean and to only get information from people you care about. Of course, LJ also recognizes that there are times when you need plausible deniability. It allows you to create a separate group of LJ folks that you actually watch (separate from your “friends” list). The subscription process is inherently a process of attention relations, not friendship.

Of course, the computation needed for directed graphs is much greater than for undirected graphs. Is that the main reason that most services require reciprocity? Even when it’s not the best mechanism for the system? Or are there other reasons why folks are obsessed with undirected graphs?

Print Friendly

10 comments to Attention Networks vs. Social Networks

  • Excellent insight. I think I can tell you why undirected graphs are more popular to social network designers than directed graphs. It’s two for the price of one. In the early stages of deployment, there is a fear from the developers that no one will use the system or that it won’t reach a critical mass to be successful. The goal is to add as many links as possible as fast as possible. When someone creates a friend link, the system gets two links and creates more hit returns on searches and more connections of connections (of connections…FOAFFF(n) – friend of a friend’s friend’s friend?).

    Of course, once the system has taken off and critical mass has been reached, this becomes a liability in the ways you discuss.

  • John,
    >The goal is to add as many links as possible as fast as possible. When someone creates a friend link, the system gets two links and creates more hit returns on searches and more connections of connections (of connections…FOAFFF(n) – friend of a friend’s friend’s friend?).

    That argument is countered by the fact that people create more unidirectional links than bidirectional links. I have 162 people in my blogroll but have a lot less people in my immediate social circle. There are about 64 people in my IM buddy list and the main reason there are so many is because I work on the IM product. Before I joined my current team there were less than half that.

  • tom

    remember the early days of blogging when http referers functioned as an implicit directed graph of a blogger’s connections and people would have tangential semi-conversations just by posting a link to another person’s site and then scanning their referer log for the url of the reply?

    i found your site after you linked to my friendster t shirt project 2 years ago. which was my own satirical way of trying to graft a performative, undirected graph of my social world onto friendster. except for the xeni jardin shirt, all of the original batch of t shirts were of random people i’d see around the neighborhood, who were all public figures in their own small ways (a dj, a writer for a chicago-centric zine, a karaoke superstar (he actually had following)). the general reaction that i got from my parody of microcelebrity was pretty much split in two – on the one side were 20-something women w/ media/tech jobs who wanted me to use their profile on a t shirt, and on the other side were a lot of random really angry comments to the effect of “if i ever see that creep w/ my face on a t shirt i’ll kill him.”
    (i have alot i could say about the squandered potential of friendster back in the late summer of 2003 – the guy who runs that site is like a cross between dr fankenstein and the great gatsby)

    this kind of strongly negative reaction towards non-reciprocal connections seems to be getting worse as the social internet becomes more mainstream. 5+ years ago it was assumed that most people reading your blog would be strangers, at least at first, and getting to know somebody from the internet via their blog or journal required forming an opinion of thembased on information gathered without their knowledge (but with their implied consent, since it was published in an open forum.) post-friendster, with people signing up for these sites bringing an established, irl group of friends, there’s this really strong taboo against contacting or even reading (“looking at”) people you don’t already know (i’ve seen livejournal comment arguments that devolve into “why are you even reading this, you dont’ know me” from the orgininal poster) for example. maybe nobody wants to admit that a big motivator for joining these kinds of sites is pure narcissistic attention and looking for sex partners? i’m not sure what, if any, kind of dichotomy i’m trying to draw here, but there seems to be,post Friendster, alot of people who join social networking sites to use them w/ their already-existing networks of friends, and specifically don’t want to use those sites to meet new people, so havining their content restricted to pre-approved users is how they’d prefer it. friendster recently introduced that feature that lets you see who’s viewing your profile and the consensus that i’ve seen on it is “that’s so creeeeepy” (which is a word that gets thrown around at alot of online behavior and seems in some circles to be the worst kind of thing a person can be, far worse than being a troll, or just being inane or ignorant). i like the “who’s viewed me” feature because i can view a friend’s profile and it’s a way of saying “hi, thinking of you” without actually having to write the note, but i know alot of people who stopped using friendster because of it.

    but – i think that now that there are millions of people using sites like friendster or myspace or flickr, or blogging services, only a small proportion (“early adopter”, “power user”, “neophile” types) actually want to have a network of complex power relations where you might link to people who don’t link back to you, and you’re comfortable with the idea of strangers reading what you have to say, and possibly even commenting on it. most people i think just want to use social websites as a way to keep in touch with people they already know, and consider the openness of the networks to be a drawback.

  • Depends on the goals of the system. I don’t think having online social networks that accurately represent “real” social networks is the higher order goal of most systems, though they would be valuable from a research perspective. If you are trying to find a date through a friend, or a job recommendation, etc., that people actually reciprocate a friendship or work together is very meaningful. In Wallop we thought about having unidirectional networks but from a statistical perspective, when examining interaction behavior, it makes sense to represent relationships as bidirectional because you want to normalize the strength of each relationship by each person’s other relationships. That is, a friendship is more meaningful if *both* parties interact with each other more than with other people. Otherwise the most active person (sometimes known as a spammer) comes up as everyone’s best friend.

  • Good thing Yahoo! 360 now supports both kinds of connections (directed and undirected.) 🙂

    Undirected graphs are good for providing access to private data. Directed graphs are good for sharing public data.

    Randy

  • Top class article. You have highlighted an issue I hope to address (in part) with Tinfinger. My other venture is called FanFooty, so I’m well versed in the needs of fans.

  • Iteration 3 signed off

    Iteration 3 was tagged and released into the wild yesterday. Right now you can see it at http://melt.staging.greenpeace.org , the code is here (Note we’re only hosting the ‘latest’ iteration at any time, so in a week or so the…

  • Hi Danah, interesting insight. I’m trying to connect your concept with how computer network works. Here, I consider Social Network as TCP which offers reliability (at the expense of speed) and Attention Network as UDP which offers speed at the expense of reliability (from the lack of acknowledgement/reciprocity). As a result, I think that Attention Network model is suitable for dynamic social software(like delicious) and Social Network model for more ‘static’ social software (like Friendster and LinkedIn).

  • Tom says: post-friendster, with people signing up for these sites bringing an established, irl group of friends, there’s this really strong taboo against contacting or even reading (“looking at”) people you don’t already know (i’ve seen livejournal comment arguments that devolve into “why are you even reading this, you dont’ know me” from the orgininal poster) for example.

    This is disconcerting to me also. It’s worrisome how many people on LJ or MySpace apparently want to track who all their ‘viewers’ are, as it would end up deterring the casual browser… sometimes I randomly follow threads of people’s MySpace or LJ friends, and I wouldn’t want to peruse someone’s postings or pictures and get a note the next day along the lines of ‘who r u why did u look at my page omg ur a old perv’ just because I followed a link to the portrayal of their public persona. There are ways for people to be private on social networks if you don’t want non-friends connecting or contacting you (friends-listing & privacy preferences) already, I’m leery of a bigbrothery solution that deters the serendipity of reading other’s thoughts.

  • The lack of direction in the links of MySpace or Friendster are only one part of the problem. I would argue that every relationship includes direction (doesn’t one person always want it more than the other?). The missing ingredient is the strength of that relationship.

    If I’m a fan of Ian Anderson, there is a bidirectional link between us. Of course, my influence on him is insignificant, in that I am but one of thousands of fans. But the impact of thousands of fans is significant. For Jethro Tull to tour, they need fans to buy tickets, and so on. So the strength of the connection from me to Ian is fairly weak when considering my sole connection.

    In the other direction, Ian’s influence on me is likely stronger, although not nearly as strong as that between me and a close friend.

    Using the Flickr example, there are family, friends, contacts and everyone else. These four levels could just as easily identify strength of relationship. But it’s not really enough. I have friends on Flickr who are little more than online acquaintences — the connection is tenuous.

    What I’d really like to see is something similar to a tag cloud, but with link entropy emphasized. Some social network analysis tools attempt this, but it would be interesting to see tools like Flickr incorporate automated relationship strength into their community.