Research On Customer Complaint Prediction Based On Operator's Operation And Maintenance Data

With the rapid development of mobile communication network and mobile Internet business,customers'requirements for service quality are constantly increasing,and the number of complaints is soaring.Customer complaints are related to products and services forward,and networks and network maintenance backward.Therefore,if the complaints and the underlying information can be effectively mined,and the potential customers who may make complain calls due to poor network quality can be located in advance,the operators can formulate effective response plans based on these,thereby reducing the complaint rate,improving service quality,and thus winning in the fierce market competition.At present,there is still much room for improvement in the field of telecom complaint prediction in the existing research methods in terms of information extraction and prediction algorithms.In order to fully exploit the value hidden in the operator's big data to predict customer complaints,and considering that complaints are closely related to poor network quality and customer behavior characteristics,this project has built a mobile Internet customer complaint analysis and prediction system based on the research of customer complaint data,poor network quality event data and network performance data.The system uses data mining technology to predict customers who may complain about poor network quality in the future time window and to locate the cause of poor network quality.The main contributions of this work are summarized as follows:Firstly,using Spark distributed computing platform,a customer complaint prediction system based on operators'big data is designed and implemented.The system introduces poor quality event data to the task for the first time,extracts poor quality features based on this,and realizes the analysis of complaint customers and the prediction of complaint customers due to poor network quality.Secondly,a complaint prediction algorithm based on stacking model is proposed.The algorithm mines effective poor quality features,uses down-sampling technology to alleviate the imbalance between positive and negative samples,achieves VoLTE voice service complaint customer prediction based on a variety of machine learning classification algorithms and designs stacking algorithm models for improvement.Through the experimental evaluation on real data of the operator,it is verified that the performance of the prediction algorithm based on the stacking model proposed in this paper is better than that of the traditional machine learning algorithms.Thirdly,the rule of multi-source data fusion is designed,and the function of mining the cause of poor network quality of the cell where complaint customers are concentrated is realized combined with the results of complaint prediction.The mining results can be used to assist network optimizers in network optimization.