Bayesian Nets based suggestions

Joel Ratsaby (jer@techunix.technion.ac.il)
Mon, 23 Nov 1998 14:53:36 +0200 (IST)

When using Bayesian Nets for suggestive tasks such as in on-line
e-Commerce applications, it seems that the classical MAP classification
mode which chooses the most probable class as the suggestion is not
as interesting as some other approaches. For instance, consider
calculating the differences in probabilities each time new evidence
is given and then ranking them and finding the event whose
probability jumped up the most as a result of the new evidence.
This DELTA (or difference) approach can sometimes discover a richer variety of
dependencies.

I would like to hear about other people's applications where such
non-MAP approaches are used for inference.

Thanks,

Joel