Implementation of several text classification systems using the naive Bayes algorithm and semi-supervised Learning
| Dataset | http://qwone.com/~jason/20Newsgroups/ |
|---|---|
| BinaryMultinomialNaiveBayes.py | Implementation of Multiclass Naive Bayes with Binary Bag of Words (B-BoW) representation |
| MultinomialNaiveBayes.py | Implementation of Multiclass Naive Bayes with Count Bag of Words (C-BoW) representation |
| TfIdfNaiveBayes.py | Implementation of Multiclass Naive Bayes with tf-IDF representation |
| EM Supervised Learning (BinaryMultinomialNaiveBayes.py) | I used semi supervised learning to do document classification. The underlying algorithm I use is Naive Bayes with the B-BOW representation |
| Filtering Technique for EM | top-k: Filter the top-k instances and augment them to the labeled data |
| Filtering Technique for EM | Threshold: Set a threshold. Augment those instances to the labeled data Thewithconfidence higher than this threshold |
I initialized EM algorithm with different percentages of the training data (%5, %10, %50). The parameter is denoted as P
The training and testing accuracies and comparison of various representations can be found in the report.