The Research And Implementation Of SMS Automatic Classification System

Aim ing at the problem of SMS automatic classification,the paper proposed and implemented SMS automatic classification algor ithm of a semi-supervised incidence matrix combined with active learning,at the same time the algor ithm constructed and implemented comprehensive classification system of more conditions combined with other factors of SMS.The paper firstly did a lot of analyses and researches for 4800 SMS which was collected in the life,and then designed a supervised learning classifier of incidence matrix of characteristics of connection strength.The classifier constructed an incidence matrix by the connection strength of the SMS key and categories.Due to the lack of domestic and foreign text corpus,which generated a result that the consequences of classification were not accurate,so the paper also introduced a semi-supervised learning algor ithm(passive learning algorithm),which could automatically revise incidence matrix in the process of classification.It could improve autonomous learning ability of the classifier and solve the problem of supervised learning which was caused by a shortage of text corpus.In order to solve a problem which init ial SMS classification error of a semi-supervised learning was gradually enlarged in the process of classification,the paper presented a way that a combination of active learning.It could automatically revise incidence matrix of classifier by analyzing the classification results.To improve the accuracy of SMS classification,the paper constructed and implemented comprehensive classification system of more conditions which combined with the telephone number and time on the basis of classification which was based on content.The system constructed a black and white list by using the phone number,and then constructed holiday list by using time.In the process of classification,it started to filter according to black and white list firstly and then constructed a new classifier of SMS which was increased an impact factor and based on content by holiday list.The paper verified the validity and robustness of the classifier by many experiments of three categories and multiple categories and experimental results of this paper were compared with previous experimental results of the algorithm.Experiment of three categories verified the validity of the algorithm and experiment of multiple categories verified the robustness of the algorithm.The experimental data showed the classifier had better performance in the accuracy and recall rate and F,including the average accuracy of three categories was 94.7%,the average accuracy of multiple categories was 88.5%.Finally,according to the SMS classification algorithm,the paper developed a mobile application which had been used in the actual project,and brought the better SMS classification and filtering effect,which proved the validity and practicability of the algorithm in this paper.