Social networking in Radiospace
Weblogs often list affiliated weblogs as a column of links which Doc Searls first named a blogroll. A couple of months ago, on my Radio weblog, I tried a variation on this idea. Rather than hand-edit my blogroll, I wrote a script that automatically reads the list of RSS channels to which I subscribe, in Radio's news aggregator, and echoes that list on my homepage. With a tip of the hat to Doc, I called this widget a channelroll.
The channelroll was a way for me to ask a series of interrelated questions:
Would other bloggers choose to reveal their sources in this unedited way?
Many have. The potential contradiction inherent in such a choice was shown most strikingly by Jenny Levine, aka The Shifted Librarian. One day in March, I helped Jenny with her channelroll script. The same day, the Privacy Digest weblog passed along an item from Jenny's blog about the FBI's ability to subpoena bookstore and library reading records.
In an item entitled the sanctity of sources I concluded that there really was no contradiction here. We cannot, and should not, be required to reveal sources. But in many circumstances we can, and perhaps should, choose to do so.
What kinds of social organization might this data reveal?
Other modes of online social interaction take place in shared public spaces -- newsgroups, web forums -- where group identity is explicit. If you post to borland.public.delphi.webservices.soap or microsoft.public.xml.soap, you are clearly affiliating yourself. Blogging works differently. Each weblog is an individual public space. Affiliation is subtle and implicit.
At the Emerging Technologies Conference, several speakers -- notably Steven Johnson and Clay Shirky -- argued that blogspace needs more explicit mechanisms of clustering and group formation. That's certainly true. And yet we need to respect the uniqueness and power of this new mode, in which groups are defined fuzzily and coupled loosely.
Reading lists imply clusters, as do backlinks. Even as we seek to formalize these modes, we should be on the lookout for new ones, and alert to the ways in which they all interact.
Can web services help us to organize blogspace?
In Radio UserLand, my channelroll script (stored in a file called subs.txt) can be deployed in two ways. I can drop subs.txt into the /Macros directory, and refer to it as <% subs %>. Or I can drop it into the /Web Services directory, and refer to it as <% ["soap://127.0.0.1:5335/"].radio.subs () %>.
The /Macros approach is the most straightforward for most people. The fact that my channelroll script lives in /Web Services, and can be invoked directly using SOAP (or XML-RPC), is for now irrelevant since my Linksys router blocks incoming traffic on all ports.
Nevertheless it seems inevitable to me that the web services fabric now being woven for business-to-business commerce will also enrich and accelerate the organization of blogspace. The data I'll present here would have been far easier to collect were channelrolls routinely available as web services. But I can imagine much more than that. For example, in a team situation, I might want to allow you not only to read my subscriptions, but also to subscribe me to sources that you know are relevant to our joint effort.
A visualization of fifteen channelrolls
Last week I collected fifteen channelrolls and did some analysis of them. One set of results can be viewed at http://radio.weblogs.com/0100887/gems/radioSocialNet.html. The Python script that contains the raw data and emits the visualization is at http://radio.weblogs.com/0100887/gems/channelroll.txt.
At the core of the script is a Python object called a SequenceMatcher, which can compare two lists -- in this case, two channelrolls -- and produce a similarity ratio. Two instances of the same channelroll yield a value of 1. Two channelrolls that have no subscriptions in common yield a value of 0.
As you'd expect, clusters emerge. The strongest correlation connects Sam Ruby, Peter Drayton, and Gordon Weakliem. Given the small samples (just fifteen weblogs, and as few as ten RSS subscriptions per channelroll), we shouldn't read too much into this. Still, I'm sure none of those three would find this result surprising.
Even more than similarities, I was looking for differences. There is a certain sameness to a lot of the blogrolls I see. Many of those first attracted to blogging share interests in software and networking. To a first approximation, blogspace today is a community of like-minded people. But we're starting to see hives emerge. Among Radio bloggers, for example, clusters of lawyers and academics have appeared.
It's useful to identify yourself with a cluster of like-minded people. It may be even more useful to locate clusters of differently-minded people whose activities complement your own. Jenny Levine, for example, is a gateway to a world of librarians who see information technology very differently than hardcore techies do. Of the fourteen non-Jenny lists, those of Sam Ruby, Gordon Weakliem, and I were, again not surprisingly, most unlike Jenny's.
For a techie crowd, Jenny is a connector into a network of people who need to use information technology in certain ways. Techies who would like to respond to those needs would do well to pay attention to them. Visible subscription lists are one of the ways in which disparate groups can seek out points of connection.
Connections are of course multiple and overlapping. In this sample, Jenny Levine and Jim McGee have a strong mutual correlation. But although I am weakly related to Jenny, I'm strongly related to Jim. Again this doesn't surprise me, as we share interests in knowledge management and organizational dynamics.
Jenny's chart shows a strong bimodal pattern. She correlates strongly with one set of lists (roughly speaking, non-techies), and weakly with another set (techies). Interestingly, this bimodal pattern is masked in a chart that simply averages the correlations for all fifteen charts:
Here Jenny falls in the middle of the pack. Joe Jennet emerges as the person who is most different from, and Paul Snively as the one who is most like, the (arbitrary) group.
What this might mean, if anything, I leave to others to speculate about. It's clear to me, though, that there are many dimensions of relatedness. Google, for example, sees me as more closely related to Jenny, and her to me, than do my subscription-driven charts. I'm sure that both measures are true in different ways.
As we narrate our working lives online, and intertwine with other working lives, the data trails we create will yield richer and more complex visualizations. The fascinating question, to me, then becomes: "How shall I design my interface?" For example, I can choose to reveal my subscriptions, or not. I can choose to reveal my backlinks, or not. How I make these choices will be determined not only by my own sense of architecture and aesthetics, but by what modes of interaction I wish to allow -- and to encourage others to experience with me.