A small model of ArXiv abstracts
April 12, 2016I’ve been working with Dorota Glowacka of University of Helsinki on a search system built on the ArXiv.org. We have a demo appearing in SIGIR 2016. Here is a model with 100 topics built using normalised gamma priors on topics (giving each topic a variance parameter as well as a mean parameter) on the 1.1 million abstracts to March 2016. The model took about 6 hours to run on my desktop.
This is a huge PNG file (2.8Mb). YOU will not be able to view it unless you:
- so load up on a big screen,
- click on the image to enter image view mode,
- then scroll down to bottom right click on “View full size” to bring it up,
- and then zoom around to view.
arXiv.org abstracts to 03/2016
Bayesian Models 
Leave a Reply