Initial Test Logs

Introducion

The CSpell uses Ensemble MEthod (by Dr. Halil Kilicoglu) as baseline for design and implmenation. The data in the paper are used to tested during the development. The source code has been improved (by Halil) after the paper submission and thus the results are slightly better than the paper. This page shows the log of test result from each stage.

Implementation Logs

• Non-Word

  Implementation	Data	Pre-Correction	Orthographic	Frequency	Context	All
  Paper	Test Set (100)	P: 0.94 R: 0.33 F: 0.49	P: 0.57 R: 0.52 F: 0.55	P: 0.45 R: 0.41 F: 0.43	P: 0.42 R: 0.38 F: 0.40	P: 0.64 R: 0.58 F: 0.61
  Ensemble	All Set (472)	P: 0.8329 R: 0.3550 F: 0.4798 T: 87 min.	P: 0.6007 R: 0.6081 F: 0.6044 T: 82 min.	P: 0.4457 R: 0.4435 F: 0.4446 T: 83 min.	P: 0.4337 R: 0.4300 F: 0.4318 T: 80 min.	P: 0.6436 R: 0.6523 F: 0.6480 T: 80 min.
  Pre-Correction
  ND Correction	All Set (472)	P: 0.8962 R: 0.4029 F: 0.5559 328|366|814 T: 0.15 min.
  Orthographic Similarity
  Dic: jazzy + UMLS 1-to-1	All Set (472)		P: 0.7182 R: 0.6388 F: 0.6762 520|724|814 T: 0.95 min.
  Dic: Lexicon 1-to-1	All Set (472)		P: 0.6983 R: 0.6229 F: 0.6584 507|726|814 T: 1.05 min.
  Dic: jazzy + UMLS + Lexicon 1-to-1	All Set (472)		P: 0.7449 R: 0.6241 F: 0.6791 508|682|814 T: 1.01 min.
  Dic: jazzy + UMLS + Lexicon (No Aa) 1-to-1	All Set (472)		P: 0.7163 R: 0.6265 F: 0.6684 T: 0.93 min.
  Add dictionary associated split
  Dic: jazzy + UMLS + Lexicon 1-to-1 split (2 Space, SW)	All Set (472)		P: 0.7318 R: 0.6302 F: 0.6772 513|701|814 T: 0.75 min.
  Dic: jazzy + UMLS + Lexicon 1-to-1 split (2 spaces, MW + NoAa)	All Set (472)		P: 0.7453 R: 0.6327 F: 0.6844 515|691|814 T: 0.73 min.

• Real-Word Included

  Implementation	Data	All
  Paper	Test Set (100)	P: 0.57 R: 0.59 F1:0.58