***** To join INSNA, visit http://www.sfu.ca/~insna/ *****
We're also working on this extensively at the Media Lab as part of the
Reality Mining project. Our system captures and analyzes each participantís
audio, extracts keyword-based topic and context information, and uses
waveform correlation / mutual information and 802.11b packet sniffing to
establish the participantís location and the other users in local
proximity. By next week weíll have 70 rigs comprised of a Sharp Zaurus PDA
(running linux), headset microphone, 802.11b wireless CF card, and a 256 MB
Profiles of a participantís typical social behavior are built over time by
extracting conversation features such as speaking rate, energy, duration,
participants, interruptions, transition probabilities, time spent holding
the floor, popular topics, etc...
Our website has some related papers along with my SunBelt presentation:
Although Iíve spent the better part of the year getting the low-level
speech feature extraction / PDA software to work correctly, the real goal
is the analysis of the extremely rich social network dataset that the
system enables. Weíve been using a variety of Bayesian network approaches
(particularly variations on the coupled HMM), but they donít seem well
suited to multi-relational data. Iím now starting to dabble with
Probabilistic Relational Models. (Friedman et al. provide a good
introduction to PRMs: http://citeseer.nj.nec.com/friedman99learning.html)
Does anyone out there have experience/insight re: working with PRMs on
social network data?
>I was wondering if anyone was aware of any work done to record human
>conversations (like at a conference) for SNA? I know there has been
>work done analyzing videos, but I was looking for other approaches.
SOCNET is a service of INSNA, the professional association for social
network researchers (http://www.sfu.ca/~insna/). To unsubscribe, send
an email message to [log in to unmask] containing the line
UNSUBSCRIBE SOCNET in the body of the message.