Word Embedding Based Multinomial Naive Bayes Algorithm for Spam Filtering

Spam messages are widely used nowadays to promote the business. In order to tackle this issue, spam filtering algorithms are used to detect and remove spams. Naive Bayes is popularly used in spam filtering. But the major drawback of this algorithm is that it assumes independence between every pair of features. As a result, features occurring in the same context are not given weightage during classification. An innovative classification method based on Multinomial Naive Bayes and Word Embedding is proposed. First posterior probabilities are calculated using Multinomial Naive Bayes. If the absolute difference of the ham and spam posterior probabilities is less than a certain threshold, word embedding is used to find out the closeness of the features in vector space. Results show that Multinomial Naive Bayes combined with Word Embedding gives better accuracy than Multinomial Naive Bayes alone.