Backchannel is a real-time view of the conversation happening in the #etech IRC channel on Freenode. We have a bot observing events in the channel, publishing them as a web-service via XML and JSON, and making them available to a visualization component built in Flash.
A conference backchannel is home to a lot of interesting conversational dynamics: additional context, participation from speakers, visitors from elsewhere in the world, sniping, and coördinated walkouts. We’re intrigued by the idea of a pointedly live visual environment that illustrates and responds to this other space.
Backchannel was built to visualize IRC communication for O’Reilly’s 2006 Emerging Technologies Conference, where it was favorably noticed by such luminaries as Information Æsthetics, Waxy.org, Valleywag and Suicide Girls. It was subsequently repurposed for the SXSW Interactive Conference, and is now attached to O’Reilly’s 2008 Emerging Technologies Conference.
Backchannel is a product of San Francisco’s Stamen Design, a boutique services firm specializing in data visualization, map making, interactive media, and creative technology reinterpretation. You might remember Stamen from such projects as Mappr, MoveOn.org’s Virtual Town Hall, and Graffiti Archæology with Cassidy Curtis. Backchannel was built by Stamen’s Michal Migurski. Our Tom Carden and Eric Rodenbeck are at the conference.
Backchannel is represented in IRC by a bot named brodmann17bot (b.17bot), named after a part of the cerebral cortex responsible for primary visual perception. B.17bot is implemented in Twisted Python, and rebroadcasts all intercepted events over HTTP. PHP has a walk-on role proxying away XMLHttpRequest security concerns.
The client software is a combination of Flash and dynamic HTML, in communication with b.17bot via XML and JSON. The initial load pulls the most recent 500 “events” (messages, action, people joining or leaving the channel) into the Backchannel Flash client, which displays all users in a circle, with a graphic indication of their participation by each name. Users who have spoken within a half-minute or so of one another are linked by threads, showing approximate conversational subgroups at any given time. A stream of events is also displayed across the top of the screen, with a slider that lets you check out the state of the conversation at any moment over the previous three hours.
Clicking on a user or an event in the stream calls up a short, browseable transcript on the right-hand side of the screen.