- Naive Bayes
- Add-1 smoothing
- 10-fold cross validation
- regular expression detecting negation words
Besides the regular Naive Bayes method, the code also realized:
- Boolean Naive Bayes
- Naive Bayes with stop word
- Naive Bayes with negation features (not, never, n't)
Accuracy:
- Regualr NB: 81.65%
- Boolean NB: 73.85%
- NB with stop word filterd: 81.10%
- NB with negation features: 81.80%
- sklearn LinearSVC
- 10-fold cross validation
- accuracy 88.45%