Bbd writeup of Borthwick et. al.

This is a review of Borthwick_1998_exploiting_diverse_knowledge_sources_via_maximum_entropy_in_named_entity_recognition by user:bbd.

I liked

• This technique allows modeller to concentrate only on finding useful features that can help extraction. ME estimation after learnign from taining data, makes sure that more useful features get more weight than not so useful features.

• Use of Viterbi algorithm (dymanic programming) ensures getting optimal solution given the selected features and weights.

I didn't like

• Compound features will really be important in the scenario where features come from multiple catagories and applied on history views. They mention that, feature selection technique they have used won't work efficiently for compound features.