Per-document Binarization is used to generate the training data of Binary Naive Bayes Classifier. Within each document, the word count is clipped to 1, that is, repeated words in the same document are deleted. Then, the words left in all documents are grouped as positive and negative respectively to the labelling of the document and their frequencies are counted.

New York University

Three major modifications are made based on general Naive Bayes Classifier:
- Binary counts: The feature is defined as a binary count indicating whether a word is present or not, instead of the word counts.
- Negation: Negation is properly handled by adding a "NOT_" prefix to the words during text normalization.
- Sentiment Lexicons: When training dataset is insufficient, using Sentiment Lexicon can improve the performance.

Binary Multinomial Naive Bayes Classifier

An on-going but a helpful book resource about NLP
https://web.stanford.edu/~jurafsky/slp3/

Speech and Language Processing (3rd ed. draft) 

Per-document Binarization

During text normalization, each word after a logical negation mark (n't, not, no, never, etc.) until next punctuation mark is prepended with a "NOT_" prefix. As a result, words like "NOT_bored" will occur more often in positive documents, and word like "NOT_interesting" will occur more in negative ones.

Negation Handling in Binary Naive Bayes Classifier

Sentimental Lexicons are used to improve the performance of binary Naive Bayes Classifier when there is not enough labelled training data. Sentimental Lexicons are dictionary of words prelabelled as positive or negative. With sentimental Lexicons, the features for each word in Binary Naive Bayes Classification can be defined as binary variables: "This word occurs in positive lexicon" and "This word occurs in negative lexicons".

Learn Before

Related