SpVar Normalization Development Notes

I. Introduction

An iterative progresses were developed to improve precision and recall on SpVarNorm algorithm by:

• Run SpVarNorm on Lexicon.2015
• Check all False-Positive instanaces
• Enhance the spVarNorm algorithm and repeat these three steps

II. Process

• Enhanced Norm to increase precision by removing genetive only at the end of term (not anywhere in the term). This is used in AMIA paper final submission.

• Test on False Positives (for increasing precision) - used AMIA final submission as baseline

  Step	Methods	Edit Distance	Sample No.	ret-rel	ret-irrel	notRet-rel	notRet-irrel	Precision	Recall	F1	Accuracy	Notes
  0	GoldStd	N/A	867,728	379,269	0	0	488,459	1.0000	1.0000	1.0000	1.0000	1 Min.
  1	Baseline AMIA-Final	N/A	867,728	305,309	3,495	73,960	484,964	0.9887	0.8050	0.8874	0.9107	1 Min.
  1.1	Genetive SpVars	N/A	867,728	303,818	1,759	75,451	486,700	0.9942	0.8011	0.8873	0.9110	1 Min.
  1.2	Dash SpVars	The False-Postive is very small (199), no enhanced algorithm is implemented.
  1.3	Space SpVars	The False-Postive is very small (41), no enhanced algorithm is implemented.
  1.4	Mixed case SpVars	These False-Postive is actually a valid (TP) due to the error in gold Standard

• Test on not to remove genitive at all in Norm

  Step	Methods	Edit Distance	Sample No.	ret-rel	ret-irrel	notRet-rel	notRet-irrel	Precision	Recall	F1	Accuracy
  0	GoldStd	N/A	867,728	379,776	0	0	487,952	1.0000	1.0000	1.0000	1.0000
  1	Norm	N/A	867,728	315,241	10,520	64,535	477,432	0.9677	0.8301	0.8936	0.9135
  1.1	Norm-no remove genetive at all	N/A	867,728	302,580	1,620	77,196	486,332	0.9947	0.7967	0.8848	0.9092

III. Discussion

• We want a Norm with very high precision (even the recall is lower).
• The recall can be improved by applying phonetic algorithm, such as Metaphone, Caverphone, etc.
• If the precision is low at the begining (Norm), it would keep going down when we apply phonetic algorithm.