Apappu 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:apappu.

Task: Exploiting diverse knowledge sources via MaxEnt in NER

Features Used: capitalization features, lexical features, style information (bold, italics, font)

Comments ---

• I liked the dictionary features, but I think they only want to address the head of Power-Law distribution.
  • I'm not quite sure what you're saying here...?

• I thought they are referring binary features to all feature classes or they may have a different notion of binary feature.

• Feature selection seems to be a heuristic selection, that might make some independent assumptions about the data. Instead, Regularization of features might work well.
  • I agree!

---