Multinomial Naive Bayes with EM algorithm

This project implements the multinomial Naive Bayes with Nigram's many-to-one assumption EM algorithm (semi-supervised learning).

The implement of EM algorithm derives from the basic multinomial Naive Bayes of scikit-learn's MultinomialNB.

How to reproduce

• Install requirements.txt
• All test cases are implemented in source.py as individual functions.

Main modules

• data/... all data processing scripts
• multinomial_model implements model,
• utilities supports some simple graph construction methods