Introduction
Dataset: Reuters-21578 Text Categorization Collection [link]
Goal: Classify the TOPIC for given news article (135 topics)
Method: SVM
Approach
Create a vocabulary from the news articles. Lemmatize/stem the
words. Eliminate the stop words. Eliminate very rare words. Use
reduced vocabulary as features. Use SVM to train the model on
different kernels. Evaluate on the test set. Try to find more way to
modify features (for example use document length or word
occurrences).