Conversation Clouds

Here’s what I mean by conversation clouds:
The cloud would be a picture of a conversation surrounding a person or a topic. The picture would show the relationships between the participants in a conversation. The densest areas would represent people who frequently cross-reference each other over time.
You can start with a participant (the url of a person’s weblog), or a search term (a word or tag) Nodes are clustered based on closeness, measured by number of links and reverse links over a period of time (comments, too, if you can measure them).
If the picture starts with a link, then that link is at the center of the picture. The picture shows the links between the first node and the other nodes, and between other nodes that are connected to each other.
If the picture starts with a word, topic, or tag search, then the cloud contains a cluster of blogs that include the term or tag in the last time period. The picture shows lines between blogs that link to each other. Unlinked blogs are thrown out.
The cloud is built from a data set over a time period; the user should be able to scale the time (conversation over a week, a month, six months) The conversation cloud would need to provide ways to navigate through conversation space. If you click on a blog, perhaps you re-center around that blog’s conversations. If you click on a tag or topic, you search based on that. You’d need to experiment with several ways of allowing browsing out from the first cloud.
This type of picture would not measure rank. Instead, it would illustrate the connections within subcommunities.
Cloud-browsing represents a pattern of blogsurfing. A reader might start with danah boyd, Stowe Boyd, Ross Mayfield.
The cloud would show in graphical form what a Technorati or Blogpulse search would — who linked to the post. And it would also illustrate the repeated links and cross-links as people reply. If you zoomed out the time horizon, you’d see some relationships become more obviously dense, with repeated patterns of links and counterlinks.
I think this sort of presentation would get more of what we’re looking for — a picture of the relationships in a community that reveals participants, both loud and quiet. The ability to browse the conversation.
The results would be more interesting than a diagram of an email thread — where participants already know who’s talking to whom. It woudn’t be particularly rankist, since webwide popularity isn’t relevant to the picture. It would let you browse to related people, or related ideas that the same people are talking about.
The next step is to test this idea, maybe with a manually drawn picture, and then with a dataset and a toolkit like TouchGraph. This seems like a good experiment to me. It could be somebody’s done this already. Or somebody’s tried this and proved that it doesn’t work. Please share if you know.
p.s. Zawodny talks about the need for content discovery. I don’t know about you, but a lot of the content that I discover comes from browsing through a conversation and finding voices that I want to keep hearing.