[jdev] what would you think of this as an SoC project?

Greg Wilson gvwilson at cs.utoronto.ca
Thu Mar 27 10:27:10 CDT 2008


> From: Greg Wilson <gvwilson at cs.utoronto.ca>
> Two grad students at U of Toronto have been working this term on a tool to 
> analyze chat conversations (IRC, IM, etc.) and cluster messages into 
> threads...
> Would this make a good project for XMPP SoC?

> From: "jehan" <jehan at zemarmot.net>
> for info, what do you mean by "analyse chat conversations"? Statistics 
> analysis? Or intelligent analysis of the content (so an AI project)?

"Analyze" == "run clustering algorithms to group individual messages
according to how related they are to each other so as to reconstruct
the logical threads of the conversation".  The clustering algorithms
themselves are statistical.  Their work does *not* include summarization
of the contents of the recovered threads, but that would be an obvious
next step.

> From: "Kevin Smith" <kevin at kismith.co.uk>
> Could you send me a copy of the journal/conference paper please? I'd
> be interested in reading about it, at least.

Nothing published yet --- this was a course project.  They will have a
screencast up on the web in a couple of weeks showing what their software
can do.  If you want, I can send references to the papers they've been
drawing on.

> From: <nicolas.verite at gmail.com>
> Check out with the mouse pointer: http://linuxfr.org/board/

Times out on me.

> An user using the web-based interface clicks on this timestamp
> in order to answer or reply to a line of chat.
> Thus by hovering all the referenced timestamp,
> one could easily read, follow, and instant view of mixed chat threads.

Interesting; Tong and Yaroslav's system doesn't require any extra
effort on the part of users, or use of new clients.  I'll point them
at it.

Thanks,
Greg



More information about the JDev mailing list