The Research Of User Classification Algorithm Based On The Regularized-naive Bayes

Recommendation system has very important applications in various situations such as e-business,information push server.It classifies training data preliminarily,then determines the user's personal preference and sends out the personalized service information based on the result of preliminary classification.This kind of system increases the degree of user satisfaction.Naive Bayes Classifier has been used in the user segmentation of recommendation system because of its good classification performance.In real life,because of the fact that the incomplete training data has the constant problems of high dimension,it leads to poor classification accuracy,bad time calculation and low anti-noise ability.For these problems,in this paper,with the help of matrix decomposition and regularization technique improved Naive Bayesian Classification algorithm,our study will propose Regularization Naive Bayesian user segmentation algorithm.The main research content is as follows:Gaussian Naive Bayes is equivalent to a diagonal covariance Naive Bayesian Classifier.During data training,we add penalty factor and Shrinkage regularization method to Gaussian density function which provides maximum likelihood estimation to prior probability,in order to reduce Naive Bayesian Classifier's over fitting and improving the classification accuracy as well.We use Ada Boost adaptive algorithm to improve regularization of Naive Bayesian Classifier accuracy.First,we set the initial weak classifier for the training sample data set.Then we obtain the optimal combination of the weak classifier weight system by iteration method.The powerful committee is made up of weak classifier,through this way the accuracy has been improved.Finally,the proposed algorithm will be applied to the user classification problem.And the related experiments will verify the feasibility of the algorithm based on Movie-leans and UCI data set.The experiment shows that optimization algorithms are better than other algorithms in accuracy and efficiency through comparing with the Traditional Bayesian,Kernel Function Bayes,Support Vector Machines and Decision Tree.This paper will introduce the fact that it is feasible that using Shrinkage regularization method and Ada Boost technology to improve that Naive Bayes Classifier.