Philgoo Han wirteup of Borthwick, Sterling, Agichtein and Grishman

From Cohen Courses
Jump to navigationJump to search

This is a review of Borthwick_1998_exploiting_diverse_knowledge_sources_via_maximum_entropy_in_named_entity_recognition by user:Ironfoot.

This paper is about named entity recognition using Maximum Entropy. Maximum entropy is a learning model based on the chosen feature space which makes appropriate chioce of features critical. In quality case including as much features as possible may help but this will lead to a significant drop in performance. This paper didn't handle the tradeoff very much but left is as a future work.

  • Good point - being able to deal with many features is part of what makes a learning method good for NLP tasks.

Leaving the feature selection method, I can find this paper has considered many promising features, of course referred from previous papers. But I still have the question what research has been done in the linguistic aspect on finding features. Are the features close to optimal?

Retrieved from "http://curtis.ml.cmu.edu/w/courses/index.php?title=Philgoo_Han_wirteup_of_Borthwick,_Sterling,_Agichtein_and_Grishman&oldid=737"