Re:Stanford PhD Dissertation Browser
by abeppu on Wednesday December 01, 2010 @02:42PM
(#1959252)
I'm a little annoyed that they used the phrase "topic distance" but it looks like they're using something which is in some cases very asymmetric (the 'distance' between CS and EE is not the same as between EE and CS). As a visualization, it's not meaningful because they don't explain what it means for two topics to be 'close'.
I'm guessing they're using something like a KL divergence between the distributions over words (smoothed?) given each topic, but that could be way off.
I'm guessing they're using something like a KL divergence between the distributions over words (smoothed?) given each topic, but that could be way off.