IPTV User Complaint Prediction System Design And Implementation Based On Spark Distributed Computing Framework

With the development of the Internet,IPTV has been rapidly launched by operators and Internet companies.However,due to the IPTV services is still in the promotion stage,and there are many application problems,especially the poor user experience.In order to improve the quality of user experience and adapt to the time of the big data,operators hope to establish a user complaint forecasting system based on key performance indicator(KPI)data collected by IPTV set-top boxes.That is,through analyzing the data of KPI and establish a complaint model,real-time prediction of potential user complaints,convenient operation and maintenance personnel to contact the operator the relevant user and timely maintenance,so as to improve the quality of the user experience,thereby enhancing the quality of the user experience.This article starts from the background of IPTV application,and combines with the KPI warning index selection,user complaints system modeling,system design and implementation,launched a series of research.The specific research contents are as follows:(1)Firstly,preliminary data preprocessing and preliminary index screening of the IPTV KPI data are performed.Then,the PCA method is used to propose a correlation index measurement method,RePCA,to inversely analyze the index data obtained from dimension reduction.The index that has the greatest impact on the forecast results of reported obstacles.Its purpose is to further reduce the number of indicators of the data based on the screening of preliminary indicators,which can greatly reduce the amount of calculations during predictive classification and processing,so that the reportable obstacle prediction system has real-time performance.(2)Secondly,According to the characteristics of imbalanced KPI data in IPTV set-top boxes,this paper proposes the EMCNE method and the SVM-KNN method for data modeling from the algorithm level.In the EMCNE method,the problem of imbalance in within-class is considered,and the spatial characteristics between test data and training data are taken into account in the ensemble rule.In the SVM-KNN method,the imbalanced of KPI data is solved by using the method of combining the spatial distance of data and the weight of the minority classes.The experimental results show that compared with the traditional ensemble rules,the two methods all improve 40% when the Performance evaluation index is AUC(Area under the Curve of ROC),and the effect of SVM-KNN is slightly better.(3)Finally,the predictive model obtained by the modeling part is applied to the Spark platform and integrated with the front pages to form an IPTV user complaint prediction system,which is convenient for operators to find fault prediction users timely and contact or repair in time.This will promote IPTV marketing and improve user experience.