Wka writeup of Sha and Pereira 2003

From Cohen Courses
Revision as of 09:06, 24 September 2009 by Wcohen (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

This is a review of sha_2003_shallow_parsing_with_conditional_random_fields by user:wka.

Comments

  • (-1) They showed CRFs to be only competitive, instead of attempting to go one step further towards an improvement (results are not very impressive)
  • (-1) They have (many?) resource constraints
    • They did not use feature selection
    • They use the complete feature set for the highest F score, but only the supported features (not clear whether this is on all other results)
    • But they say that the complete feature set may in principle perform better; if they could just do that!
  • (-1) In parameter tuning, it would have been nice if some work was done on explaining F score fluctuations with log-likelihood
  • (-1) More on "Labeling Accuracy"?
  • (+1) Nice review of CRFs and training methods
Retrieved from "http://curtis.ml.cmu.edu/w/courses/index.php?title=Wka_writeup_of_Sha_and_Pereira_2003&oldid=814"