Skip to content

Latest commit

 

History

History
20 lines (15 loc) · 988 Bytes

METRIC.mkd

File metadata and controls

20 lines (15 loc) · 988 Bytes

#Sentiment Analysis METRIC ##Feature Extraction Currently, features with unigram and bigram mixed bring the best performance in general measured with precision when cross validation and manually tagged data set.

##Test Data Description 490 stats manually tagged, of which 257 are POSITIVE, 233 are Negative.

##Algorithms Metric (bi classes)

  1. NaiveBayes 78.8% (386/490) (mixed, 10-cross: > 96%(with icons) 70.51%(without icons))
  2. SVM 86.1224% (422/490) (mixed, parameter description: c: 0.0004, B: 1. 5-cross: 98%(with icons). 74.8%(no icons))
  3. LR 82.6531% (405/490) (mixed, B: 1. 5-cross: 98%(with icons). 76.63%(no icons))
  4. Max Entropy 88.37% (Only unigram, Iteration: 4.)

##Algorithms Metric (tri classes)

  1. NaiveBayes 10-cross: 88.45%(with emoicons) 71.88% (no icons) test
  2. SVM 5-cross: 99.9714%(with icons) 82.1212% (no icons) test 73.9305% (553/748)
  3. LR: 99.9818%(with icons) 83.6106% (no icons) test 75.5348% (565/748)
  4. Max Entropy test