Naïve Bayes regression model and its application in collaborative filtering recommendation algorithm

Abstract

Aiming at the problem that the general Bayes classification model cannot be applied to regression prediction, a Naive Bayes regression model is proposed. Firstly, in order to simplify the complexity of model, attribute values and decision values are discretised from continuous value to discrete value. By summing the probability of Bayesian classification and using the mathematical expectation value as the regression value, the process of regression problem is turned into a classification problem and the Naive Bayes classification model was modified into Bayesian regression model. The difference of input attribute values, output values, application scope, requirement of output values, operation to obtain output values between Naive Bayes classification model and Naive Bayes regression model is compared in detailed. Secondly, the Naive Bayes regression model is application in collaborative filtering recommendation. This study identifies the user and the item as independent attribute characteristics, while rating as the classification category. And the attribute values and category values are discretised to simplify the complexity of Naive Bayes regression model. The experiment results on Movielens-100k, Eachmovie and Jester data set show that this new method has high success rate and efficiency.