Performance Tests for Ensemble

I. Introduction

Performance tests on the test set are conducted on the Ensemble Spelling as the baseline to compare to CSpell.

II. Setup

• Program:
  ${C_SPELL}/SpellCorrection/bin/runSpellingAllData
4 (CSpell data - NER)
  3, 4 (nonword, real-word)
  4 (methods)
• InData:
  ${C_SPELL}/SpellCorrection/CHQA_SpellCorrection_Dataset/ResultCSpellData/
• OutData:
  ${C_SPELL}/SpellCorrection/CHQA_SpellCorrection_Dataset/ResultCSpellData/LinearWeighted_nw_OUT_4
${C_SPELL}/SpellCorrection/CHQA_SpellCorrection_Dataset/ResultCSpellData/LinearWeighted_rw_OUT_4

  Backup on:

  • ${C_SPELL}/SpellCorrection/CHQA_SpellCorrection_Dataset/ResultCSpellData.baseline
  • ${C_SPELL}/PostProcess/data/Test/NewTest/TestData/9_Baseline/Offical/*

III. Performance Results

• Non-word Only GoldStd

  Methods	Revised GoldStd TP|Ret|Rel Precision|Recall|F1
  4. Ensemble	559|966|974 0.5787|0.5739|0.5763

• Read-word Included GoldStd(Only Ensemble option works for Real-Word)

  Methods	Revised GoldStd TP|Ret|Rel Precision|Recall|F1
  4. Ensemble (NW)	560|966|1178 0.5797|0.4754|0.5224
  4. Ensemble (RW)	520|810|1178 0.6420|0.4414|0.5231

The results of non-word and real-word options from Ensemble seems do not have too much difference.