Concept - Document Classification

Definition (s.41): Given a collection of labeled documents (training set); find a model for the class as a function of the feature values; goal: assign previously unseen documents a class as accurately as possible.

Applications (s.41): topical classification of news stories / web pages; SPAM detection; Concept - Sentiment Analysis.

Methods commonly used for text (s.41): naive Bayes; support vector machines (SVMs); recurrent neural networks (RNNs), e.g. LSTMs; KNN or random forests may also work.

Tricks (s.45): select top-k words by correlation with the label; use Concept - Sentiment Lexicon for sentiment tasks. Labeled data is expensive; web reviews with star ratings serve as labels (s.43).

Source: 01 Intro - Chunk 02 (s.31-48)

Related

Concept - Pattern Discovery, Concept - Document Clustering, Concept - Feature Selection