Jan 1, 2009

So many people, So little time

There are many interesting discussions happening on Twitter in the last few weeks. I have been paying attention to the conversations about how to measure social media influence/rank/authority. Many people trying many metrics. All of the current measures seem to have their issues... one gives too many high scores, another spits out way too many around the median, and still others are easy "to game."

IMHO, the best metric so far is Retweet Rank -- it is not perfect, nor complete, but it gets at a key aspect of Twitter -- what is interesting, what are people paying attention to, what is useful and who is sharing all of that. If someone re-tweets what you posted/tweeted then that is a vote of attention/quality -- if many people re-tweet you, or the re-tweet the person that already re-tweeted you, then better yet! In this sense Retweet Rank follows the example of Google's very successful PageRank algorithm -- people point to web content they find interesting/useful/valuable.

In a recent exchange someone asked me, "Why do you follow so few people on Twitter, yet many people seem to follow you?" I answered, "Simple, it takes effort and time to follow people, but much less time & effort to have many people follow me." This reveals a typical constraint in all of our social networks -- we only have so much time/energy/attention to devote to people in our lives, we cannot be connected to everyone. Even on the internet we must choose where we pay attention.

The diagram below shows an individual's social network. The person in the middle [red node] has a set [green nodes] of friends/family/colleagues or people they follow on Twitter.

Each of those friends/colleagues/follows also has friends/colleagues/follows. We build a fractal-like network map -- each green node above has a network map like the red node above -- self-similarity through all levels. The 2 step network of the red node above may look like the diagram below -- a social spore? Click on picture below to see detailed structure.

Each of the social circles seem to be isolated from each other. Looking at this picture of a person's social network, we think we need to follow at least one person in each social circle/cluster/community to know what is being said in that group.

But the picture above is misleading -- it forgets one important aspect of all social networks. Friends of your friends are often friends. Colleagues of your colleagues are often colleagues. Those ties are missing from the simple picture above. The map below includes the links that exist between and amongst friends and colleagues. Click on picture below to see detailed structure.

Our software finds the emergent patterns in the data, but the massive interconnectivity between friends and colleagues does not leave us with a pretty picture like the simple social spore above. Click on picture below to see detailed structure.

So, if many of the social circles above are already interconnected do I have follow an individual in each social circle/community on Twitter? Probably not. The trick is to find the people that reach many social circles and follow them. Of course, we need to find more than the minimum of people to follow -- you want some redundancy in your network so that there are multiple paths to places of interest for you. Finding these key nodes is what social network analysis is all about.

And this is why I follow so few people on Twitter! For the time invested, I want maximum return. I use the redundancy of connections, between the many social circles I am interested in, to my advantage. I follow a select group of people that give me the same access as following someone in every group. Follow the few to reach the many!

Because I have chosen them carefully, I want to actually read the tweets of the people I follow. A small part of my "following network" is always in churn, but the number of people I follow on Twitter never exceeds 100 [currently I follow about 70]. Those who follow thousands of people readily admit that they can not read the fire hose of tweets they get every day.

Strategically I am building a small, yet efficient, group that reaches out into the many diverse information pools I am interested in. I know I am finding good people to follow on Twitter by the number of great exchanges that emerge on many topics. Think before you follow, use your time and ties wisely!

UPDATE #1: Looks like Stowe Boyd advises to keep "Following" count >= 100. Hat tip to Robert Patterson. I will usually add new people to follow and then may keep them, or substitute them for someone else after a week or so. I tend to have a core of 50 people that do not change much, and then another ring of 20-30 that churn. People that stay in this later group eventually become part of the core -- which grows slowly over time.

UPDATE #2: Another reason not to auto-follow people was revealed by the recent Twitter SPAM and HACK attacks. People were spammed via the "direct message" [a.k.a. "d m"] protocol in Twitter. A direct message in Twitter is a private correspondence between two people on Twitter that are following each other. A dm is like a short private email. You can only get a dm from people you follow, and if you auto-follow, chances are you will be spammed. Don't be a promiscuous follower, practice safe following!

Password theft is going on in Twitter. This is how people are breaking into Twitter accounts. What may be happening is password harvesting via the many Twitter rank/authority/influence applications that are so popular now. Many ask for your Twitter username AND password. I avoid those apps and just visit apps like Retweet Rank that only ask for your public user name and not your private password.


  1. Fascinating pictures! I think the right amount of people following to also depends on the nummber of posts they do

  2. As long as there are popular people (and robots) who automatically follow anyone who follows them, the social network graph you illustrated will be bogus. I'm finding interesting comments and pages by tracking the movement of URL citations in the social network that I'm close to. See http://www.nickarnett.net

  3. Nick,

    The maps represent how real social networks look -- we all living in overlapping social networks, yet we often focus on our own little world/cluster of immediate friend/family/colleagues.

    This has nothing to do with bots or automatic re-follows! In fact I am arguing *against* auto re-following in the post.

  4. My point was that maps of Twitter follower patterns do not look like real social networks because of automatic following.

    Your approach to choosing who to follow is valid, of course - you probably know it has been used in intelligence and law enforcement for a long time. I'm only quibbling with your implication that mapping follower relationships on Twitter reflects reality.

    (I know what real social networks look like. I have several patents for methods of analyzing online social networks. Been doing this for years.)

  5. Nick,

    One last time... the maps are *representations* of what connections may look like between various social circles -- on-line or off.

    I have been social network analysis for 20+ years, you can also visit my web site to see actual social network graphs from real companies and real communities.

    Actually, you and I are in violent agreement! I have been pointing out that FB, LI, and most social networking sites do not have anywhere near an *accurate social graph* of their members, with a big reason being the the network noise created by "link whores" and those who accept all friend requests.

  6. Awesome write up Valdis.
    Thanks for appreciating Retweetrank.

  7. Sorry if I seemed obstinant... it just seemed like you were suggesting that if one mapped the follower relationships on Twitter, you'd get something that resembles the real world.

    I know you've been doing this for a long time, I've been "following" your work for years, on and off.


  8. OK Nick, we agree the maps were metaphors -- just like in this example.

    Yes, mapping many of the current on-line social networks does not give us an accurate picture of the actual social network of that person. But understanding basic patterns of social networks helps us build better ones -- even if we do not have a perfect map of the 'as is' network.

  9. Excellent write up and love the visuals of the network.

    I agree with you that you don't have to follow everyone in the world to have a powerful network. I also agree with your approach of finding key nodes across different networks.

    But, with tools like Tweedeck and Twirl that you create groups of twitter streams - can you have the both of best?

    I am following 3,000 people. I don't read every single tweet they write they write every day. But I take a dipping approach into the big network - check in on the stream for like five minutes each day and see if there is a random act of kindness - offering some information or whatever.

    Is that of value? Yes! I discovered a tweet from one follower about an Einstein quote that provided the inspiration and conceptual framework for a webinar on best practices in social media for nonprofits -

    So, the dipping is like sharpening a pencil or way of finding some inspiration or a different way of thinking.

    Then, I also take the approach similar to what you outline - but I use Tweetdeck which is an IM client where you can group Twitter users into subgroups and then read those streams. I spent 10-15 minutes and read those streams bit more indepth. These streams provide me with excellent information and ideas as well.

    And, of course, I spend another 5-10 minutes - as needed - asking my large number of followers questions or crowdsourcing ideas.

    So, I'm discovering new people but building strategically across my networks - investing maybe 20-30 minutes a day. For me, it's been of value.

  10. Totally agree Valdis. I always keep Dunbar's number in mind when I come across a request on Facebook, LinkedIn, Twitter etc. My question is always 'how will a relationship with this person be of mutual benefit commercially, intellectually, socially?' Answers to that question usually easy & keep me safely within Dunbar's number!

  11. This will get even more interesting (and useful?) when tools become available that allow the type of social graphs depicted to become more dynamic, information dense and navigable representations of the social networks. There are a number of parameters associated with the appearance of Nodes and Edges that could possibly be manipulated to provide extra information about the underlying social network, but I think even more important would be the interfacing back from the visual graph to the source. I got the sense this is getting closer after exploring a bio-informatics tool "Cytoscape", but level of complexity still way up there.

  12. I know, you are aware of it. But I leave my comment anyway. Compared to many visualizations of scale-free networks the pictures look pretty boring. Is this due to the fact that the following/follower model is also pretty boring (not to call it fascist).
    A good idea would be to follow "relations" over diffent media. From my own observations I know that I have strong overlappings in different media like Xing, LinkedIn, FriendFeed, Blogger, TikiWiki, etc.

  13. technically the 1st network graph is not fractal == self similar because it is not scale-invariant but a low coordination number lattice (eg Bethe lattice <4 coordination n.)

  14. Yes, MgP you are correct... only the 2nd graph is an intentional fractal. The 3rd and 4th graphs are built upon the 2nd.

    Wow, looking at all the comments to this post, this blog has a VERY intelligent readership! Thank you all!

  15. What Travelwriter said. Some people don't post often, others post a lot. While 100 would be nice, I have found it worthwhile to follow nearly 400. But it is a stretch, and closely related to my job. I could definitely prune.

  16. I joined Twitter a little over a month ago and have been studying it intently. In graduate school I was studying chaos theory and have felt that Twitter is a great example of it in action. Managing change on the edge of chaos was the theme as a training manager in looking to create a learning organization. Now I am a Realtor in a very chaotic market and this post completely supports the idea of giving up control but knowing that there are patterns underneath it all. I am so glad I found this post--Fractals, the rule of 150 all came flooding back as I make sense of my experience on Twitter.

  17. I very much agree with the general thesis based on our experience. We might consider two other open issues:

    (1) We are viewing early stage analytics through early development of social portals that are designed with limited, or no clear purpose. There is another evolution that will better enable multiple forms of network behaviors, and might alter how we view the numbers - although not our "supply" of attention.

    (2) It seems to me that different networks require different levels of involvement depending in part upon that purpose. If we consider that each individual has a total supply of attention, and that the mix of networks in which they are active varies, as well as the attention/involvement requirement of each network, then, the connection numbers start to vary as well around some distribution yet to be defined. In other words, on Twitter, 75 followers might be appropriate for Valdis, while 100 followers might be optimal for Stowe, etc., Each person has an optimum which varies depending on what they are trying to achieve in their "portfolio" of network involvement.

    In a publishing model, I understand why Kawasaki would have 20,000 followers, but he again is acting more like a publisher with an occasional interaction. He probably interacts more than most with that level of followers.

    But my prediction is that next wave network platforms will enable intermittent engagement in business services that are less time intensive for the network member, and that support "weak links" between members, but strong links to content.

    Theory of bounds has to be true. Sampling behaviors that enable dynamic refresh of networks must also be present.

    One other thing that you touched on - as you become familiar with followers, or network members, doesn't your transactions cost go down based on your growing trust based on relevancy etc., Meaning that the carrying capacity of the "pipe" actually does grow over time limited by our ability of absorption?

    The other issue that is not being addressed is the qualitative difference between surface communication in networks (Twitter) versus communications requiring "deep" reading and analysis. They have entirely different cost characteristics, that again affect draw upon total supply of attention.

  18. We need to remember the difference between followERS and followING-- those we follow.

    In the post I was focusing on those I follow -- I can only pay attention to so many.

    My followERS are not so limited. Unless a conversation starts with many followERS, I am not actively involved. I'm posting stuff, having interactions, if they like they stay, if they don't they leave.

    Many followERS are like lurkers on listservs & online groups -- we know they are there and paying attention to what the group produces, but they do not require our explicit attention, until they speak up. More attention is required for followING then to followERS.

  19. I am interested in (eventually) seeing visualisation of tweets on topics and RTs, to seek (I think) the revealing fractal patterns of human verbal (text) activity in networks ...

  20. What bothers me with retweetrank is that many of the retweets that I see are to Tweets that are in-turn linking to someone else's content, typically in a blog. The linker gets the retweetrank love, while the original author gets no recognition. Okay, the retweeter does provide a valuable network function, e.g., as you indicate, reducing the number of people you need to follow -- connecting what might otherwise be isolated clusters in the network. Would it be better to also track source-content URLs...which could be either a Tweet (if no link within) or blog and then show side-by-side ranking for these to give a more complete picture of who is putting value into the network, at least as valued by the masses?

    An additional noise factor comes from retweet behavior being under reported as some, (familiar with the RT convention or not) including myself, discover content via Twitter, perhaps put it in a queue to read later, then when actually reading, digesting, and confirming the suspected value will link to the original content in their own fresh Tweet; versus broadcasting into their network as a retweet. This behavior generally fine in my book (especially given a 140 character limit) as I'd rather know the additional thought/comment (i.e. value-add) of the person I follow than to know how they found the link from someone else...especially as I find reading through RT's so visually distracting. Admittedly the downside of this behavior is it shields from discovering new users (the one being retweeted) to follow which are closer to the source, and potentially more valuable in your network. Overall, I think what I am reaching for is closer to the Delicious "via:" pattern attribution versus what can feel like reading a relay-station log.

  21. When I originally started using Twitter I applied similar rules to the number of people I followed.

    Numbers of followers increased slowly with time -- definitely wasn't a situation where I decided today I will add 500 people to my account. People add me and I choose to follow most people unless they obviously are spammers, their interests are too different from mine or they mainly use twitter for announcements and don't engage in conversations.

    Time wise there isn't necessarily much difference between 100 to 1000. Weird as it sounds I'm considerably more time efficient following more than I was with less. What changes is the nature of the conversation, less than 200 feels considerably more intimate but more than 200 provides more diverse idea exchange plus greater chance of faster assistance.

    Also gets back to why/how you use twitter. For me more people saves me time and makes me more effective at my job.

    But to be perfectly honest, if I was following a low number of people and had it set to all @replies I definitely wouldn't follow me because I tweet too much. However apparently some of my followers find my tweets funny.

  22. The mesmerising images are no doubt beautiful depictions of tracking, linking and complex de-following activities. But there seems to be an unquestioned assumption that
    i) every social media user is a homo economicus and
    ii) quantity preceeds over quality.
    Luckily, qualitative research makes use of the quantitative aspects but would not neglect the meaning-making processes which are behind those graphs. Each time we check figures, stats and curves we might feel more or less a virtual celebrity. However, the contribution made to personal growth (cultural, social and even economic capital) by comments similar to offline conversations is hardly understandable by relying on these maps only. No matter how pretty they are, and how keen we remain on measuring our status in the ever more crowded virtual worlds. Social network analysis is only one way to understand certain aspects of social circles...unfortunately a fairly one-dimensional.

  23. We agree Britta... my whole point was that I am reducing quantity to focus on quality.

    Social network analysis [not the exact numbers in this case, but the thinking behind it] allow me adjust my "quantity" so that it maximizes my "potential quality".

  24. Valdis, I've pursued opposite strategies for LinkedIn and Facebook. On LinkedIn, I only seek or accept connections from people I know, have worked with, and would recommend; or people whose work I admire and respect (and I don't seek links from people like that whom I don't know personally, but I do accept links from them when offered). The gray area of "attended the same conference" or "subscribe to the same mailing list" has led to some awkward moments where I've refused to link to someone who was probably perfectly nice. This is because I want my LinkedIn network to mean something and thus far that has served me well.

    On Facebook, a network I viewed initially as a toy and a subject of study and later a competitor to my employer, I have been much more "promiscuous" in my liking. I still try to avoid inane, vapid connections, but I will connect to someone on Facebook on a much slimmer pretext. This has led to me occasionally, in real life, warning my friends *not* to use connection to me on Facebook as a measure of credibility.

    Essentially, I did not value the meaning of my social graph on FB and, while trying to avoid spammers and scammers, I've been liberal about connecting to see what happens.

    This turns out not to have been such a bad thing, since FB's greatest charm seems to be its very large user base and the resulting "reunions" its fostered for me and others (with old grade school classmates, etc.).

    Because Twitter pushes to my attention I do try to be picky, but have let my follow list grow beyond my ability (or desire) to keep up completely. I prune every now and then and it tends to be about 2/3rds of my followers. I could probably use more churn because I like the discovery factor and don't want to use Twitter simply as a chat tool to reach my existing buddies.

  25. Quick technical question: how did you scrape the Twitter data?

  26. 25 comments? What happened to that perfectly quiet blog of yours?

    I'm not sure the number of follower, be it 100 or more, is as important as the number of daily updates. And I'm quite sad, two years after the company has proven how much they rely on open innovation that no one has implemented better clients that woudl rener that list less relevant, alloweing users to filter both by list of friends and by keyword: remove all updates from an conference that some of your friends attend but you don't care about; include replies, mentions or tags in the same stream; include tweets of a sub list only if it has been favorited or re-shared so many times; etc.

  27. nice work! I figured you would be interested in a similar talk given at TEDtalks> http://www.ted.com/talks/lang/eng/nicholas_christakis_the_hidden_influence_of_social_networks.html
    seems like Nicks metric and yours are quite similar.