Spelling Variant Patterns - Test on Lexicon-LRSPL

Norm, MES, and ES are used in a sequential order to retrieve the most spelling variant groups. This model is tested on Lexicon Spelling variants (LRSPL) for the recall. This is the inital test on this model. The results are shown as follows:

• Results:

  2014

  Step	Methods	Edit Distance	Combined No.	Total Groups	SpVars Groups	Single Groups	SpVar No.	Group (Recall)
  0	Lexicon.2014	N/A	0	N/A	N/A	0	249,231	100.00 %
  1	Norm	N/A	126,708	122,523	92,415	30,108	219,123	87.92%
  2	MES	2	14,571	107,952	104,157	3,795	245,436	98.48%
  3	ES	1	1,418	106,534	105,201	1,333	247,898	99.47%
  4	MES	3	145	106,389	105,308	1,081	248,150	99.57%
  5	ES	2	431	105,958	105,618	340	248,891	99.86%
  6	MES	4	37	105,921	105,647	274	248,957	99.89%

  2015

  Step	Methods	Edit Distance	Combined No.	Total Groups	SpVars Groups	Single Groups	SpVar No.	Group (Recall)
  0	Lexicon.2015	N/A	0	N/A	N/A	0	260,431	100.00 %
  1	Norm	N/A	?	126,675	96,216	30,459	229,972	88.30%
  2	MES	2	14,773	111,902	108,092	3,810	256,621	98.53%
  3	ES	1	1,428	110,474	109,144	1,330	259,101	99.49%
  4	MES	3	144	110,330	109,252	1,078	259,353	99.59%
  5	ES	2	429	109,901	109,563	338	260,093	99.87%
  6	MES	4	37	109,864	109,592	272	260,159	99.90%

• Future Work:
  • Only recall are tested (because the results are used for SpVar Matcher). Should try Lexicon (include no-spVar) to check precision and recall and find the optimum point.
    => The PRF model should be established for:
    • finding the optimal processes of this model
    • a measurement index when enhanced SpVar matcher and its componenet
    • Could be a short paper
  • Try different order in step to gain the best results (precision and recall)