Research of a Spam Filter based on Improved Naive Bayes
Ye Yuan, Shouzheng Li, Yuanyuan Wang, Chao Liu, Weimiao Feng, Min Yu · 2016
In spam filtering filed, Naive Bayes algorithm is one of the most popular algorithms, but its conditional independent assumption makes its reliance on training sets of sample space distribution.In order to improve the accuracy rate, sample space became so complex, resulting in the algorithm of time complexity is increased and the internal stability is poor.In order to solve above problems, this paper proposes a modified using support vector machine(SVM) of the Native Bayes algorithm :SVM-NB.First,SVM constructs an optimal separating hyperplane for training set in the sample space at the junction two types of collection,then according to its similarities and differences between the neighboring class mark for each sample to reduce the sample space also increase the independence of classes of each samples,finally using Naive Bayesian classification algorithm for mails.The simulation results show that the algorithm reduces the sample space complexity, fast to get the optimal classification feature subset,effectively improve the classification speed and accuracy of spam filtering.