Analysis And Forecast Of Power Customer Complaints Based On Data Mining

With the further development of the Electric Power System Reform,the power sales-side is opening up,number of business entities is increasing,and competition in the power market is becoming fierce.At the same time,consumer's awareness of service and safeguarding legitimate rights and interests is awakened.Their demands are diversified and personalized.Service quality becomes a standard for users to measure value of products.Service marketing has come to be the first choice for enterprises to create competitive advantage.On the other hand,companies have accumulated a large amount of detailed business data,which is large-scale,diversified and timely.Therefore,in the context of the big data era,power companies use data mining technology to analyze business,improve services and optimize management so as to give impetus to power marketing and corporate competitiveness,which has been the consensus.This thesis took the State Grid Corporation of China(SGCC)customers as the research object.Based on the historical sheets data in 95598 Customer Service Center,it applied data mining and information technology to research customer complaints.First of all,the characteristics of the data were investigated.The mathematical statistics and grey relational analysis were used for the structured data to analyze the distribution characteristics of complaints.The text mining technology was applied to the unstructured data to find out underlying causes of complaints.There is another typical problem in SGCC's customer service,defined "Repeat Call",for which an automatic identification scheme and extraction process were designed.The design was based on Chinese word segmentation and similarity algorithm.The recognition rate of the algorithm is verified.Besiedes,the thesis also studied the business association rules of Repeat Call.Secondly,the thesis predicted customer complaint behavior in the view of quantitative and qualitative.The former established a one-dimensional time series model for the total number of complaints in one month.The latter,based on the text content described by customers,modeled Naive Bayes and random forest classifiers,which could evaluate the likelihood of that single calling is customer complaint.Finally,a closed-loop management and service improvement system that combined all above results was proposed.The system contained early warning function and service promotion strategies.According to the SGCC's organizational structure,its architecture and operation process were designed detailedly to achieve the goal of lowering complaint rate and improving service quality.Combined with the research approaches and conclusions,the thesis chose C#,R and Python for coding and design analysis software,which can automatically analyze identify complaints and Repeat Call sheets.The software can also output service promotion strategies.Consequently,the research results formed an application tool.It is being used to assist the SGCC's staff to carry out their work.