Apart from manual classification, another approach to text classification is machine-learning based classification. This type of classification is a supervised learning because the set of labeled data serves as training data for the machine. This means that manual classification can't be completed eliminated. We need some annotated data in order to train our machine.
Here, for text classification, 20 newsgroups dataset is used. This dataset has 20 classes, but I have demonstrated using only 5 classes.
There are many methods for text classification. Of those,multinomial Naive Bayes and K-Nearest Neighbour are used here.
Feature selection is the process of selecting a subset of the terms ocuuring in the training set and then use only this subset of features for text classification. Here, TF-IDF based feature selection and Mutual Information are used. The two feature-selection techniques are applied to both the classifiers.