1 Introduction
Data mining brings together content of statistics, artificial intelligence, database and other disciplines, is an emerging cross-discipline. This discipline is designed to help people find valuable information from massive data, and currently in business is just starting. Domestic increasingly fierce mobile communication market competition has promoted the mobile communication operators to reduce operational costs, providing differentiated customer services, and data mining technology can help operators analyze customer consumption behavior, identify customer characteristics, assist operators Marketing and customer service.
2 data mining technology overview
"Data mining contains a series of techniques intended to discover useful and not found in data sets." [1]. Specifically, data mining is a process discovery process, which is based primarily on statistics, artificial intelligence, machine learning and other technologies, highly automatedly analyzes data, and makes a summary reasoning, and digs potential patterns, In the future, predictions are predicted to assist decision makers to assess risks and make correct decisions. Data Mining and On-line Analytical Processing (OLAP) are analyzing tools. Online analysis processing, as a verification analysis tool, "more dependent on user input issues and assumptions" [2], so that users can quickly retrieve the required data, and data mining can automatically discover mode hidden in data Pattern.
In practical applications, data mining is mainly used in the following ways:
(1) Related Analysis and Regression Analysis: Related Analysis The main analysis variables are closely related to the closerness; and the regression analysis is based primarily based on the appropriate dependencies between the observation data establishment variables. Related analysis is the basis for regression analysis.
(2) Time series analysis: Similar to related analysis, the purpose is to excavate the contact between data, but time series analysis is more focused on the causal relationship between data in time.
(3) Classification analysis: Classification analysis first gives each observation, then examining these marked observations, describing the features of these observations. This description can be a mathematical formula or model that utilizes it to classify new observations. Commonly used typical classification models have linear regression models, decision tree models, rule models and neural network models.
Cluster analysis: Different from classification analysis, the input of cluster analysis is a set of unclear records, and the purpose is to reasonably divide the record collection based on certain rules. Cluster analysis and classification analysis are a set of mutually reversible processes, so algorithms applicable in many classification analysis are equally applicable to cluster analysis.
3 Application of Data Mining in Mobile Communication
At present, mobile communications operators have business operations systems, customer service call centers, business acceptance websites, and other customer information systems, which have the following characteristics:
(1) Each information system is compared to a number of information, numerous customer data, marketing data, and accounting data in different data formats and access methods in different systems, and forms a wide range of information. Island, there is redundancy and inconsistencies in each information, and the data must have a single view (Single View) during data mining.
(2) These systems are online transaction processing, ON-Line Transaction Processing, OLTP, which dealt with online transactions in real time, and cannot adapt to data mining to apply large-scale, frequent retrieval and query operations. Therefore, the premise of data mining is to establish an enterprise-class customer information data warehouse, which can accumulate customer data from different online transaction processing systems to provide a correct, complete, and single customer data environment.
3.1 Establishment of data warehouse
As the basis of data mining, the data warehouse is different from the traditional online transaction processing system, which has theme-oriented, integrated, not updatable, and varying characteristics. As the original data source of the data warehouse, each online transaction processing system provides information on the basic data, customer call list, customer bill, customer contact history. Data warehouses processes these interface files via ETL processes (extraction, conversion, and loading), and stores and manages these customer data in different subject domain organizations. Through the data warehouse interface, the customer data in the data warehouse is online analysis and data mining. The architecture of the entire data warehouse is shown in Figure 1. It is mainly composed of data sources, enterprise data warehouses and decisions. 3.2 Topic Definition of Data Mining
After establishing a business-level customer information data warehouse, data mining can be performed based on this data warehouse platform. However, before the data mining work, you must clarify the issues that the data mining needs to be solved and the scheduled goals needed. It is only the direction and meaning of the data mining in the premise of the target clear definition. In this paper, the following topics are defined as the target of data mining in the characteristics of mobile communications operators market operations.
3.2.1 Customer behavior analysis
Using classification analysis and cluster analysis methods to analyze customer call behavior, it depends on the characteristics of customers in consumption habits, lifestyle, social contacts. The fundamental purpose of customer behavior analysis is to divide customer bases in different characteristics, and operators can perform different marketing activities and customer services for different customer bases. Typical applications in customer group division is the marketing of a certain mobile service for the consumption characteristics of a customer base.
3.2.2 Discount Policy Simulation Prediction
Offer is a very important part of marketing, and the unappropriate strategy of preferential strategies will often get an appropriate market effect. The preferential strategy simulation prediction is the effect of the preferential strategy implementation through the established customer behavior model simulation customers to predict the effectiveness of the preferential strategy. Through the simulation of the preferential strategy, the success of the preferential strategy can be predicted to adjust and optimize.
3.2.3 Customer loyalty analysis
Customer loyalty analysis focuses on the establishment of customer value models by analyzing customer consumption amount and accounting payment, thus obtaining customer value and closing tendency. The customer is the value of operators. By analyzing customer loyalty, targeted customers with high quality services, customers who have a tendency to network tendencies in time to improve operator market share, reduce marketing costs It is very useful.
3.2.4 Anti-fraud analysis
At present, the most serious problem facing mobile operators is the arrears problem, with a large part of the fraud, so anti-fraud consumption has become the key to the development of mobile communication. By analyzing the multi-dimensional analysis, cluster analysis, and isolation point analysis of customer data, customers can establish a customer fraud consumption model, which can effectively monitor customer consumption behavior, and alarms to meet the consumption behavior of fraudulent consumption models.
3.2.5 Competitor Analysis
The mature market is inevitably a competitive market. Interconnection between different operators customers is the most basic premise, so through the behavior analysis of customers and competitors customers, the model of competitors operate and customer service can be established, such as competitor customer development model, pass The use of these models can formulate effective market strategies.
3.3 Process of data mining
Under the premise of data mining targets, data mining can be performed based on the established enterprise customer information data warehouse. This article will follow these steps: problem definition, data preparation, data exploration, establish model, model inspection, model application, and return analysis, where problems have been completed in the above topic definition part.
After the problem is defined, you need to create a data mart as a data mining and analysis, generally extract data subset related to problem in data warehouse as a data market. Sampling techniques such as random sampling, isometric sampling, layered sampling, and classification sampling during creating a market; through the deletion of data and the magnification of small probability events, The characteristics and regularity of data in the city are more significant.
During the data exploration, the characteristics of the exploration data are displayed by multi-dimensional analysis and visualization, and some new variables are generated by adding deletion of data or generating some new variables to generate some new variables in accordance with existing variables; wherein data statistical characteristics Analysis plays a very important role in the process of data exploration. Establishing a mathematical model is the core link of data mining work. At present, there is a common modeling method with neuronal network model, decision tree model and regression model. Which method is specifically used in data mining depends on the characteristics of the data market and the needs of the need to implement, in practical applications, it is often comparison and synthesis for multiple modeling methods. During the modeling process, the data is hierarchically use the data and verification data, and the training data is mainly used to solve the model parameters during modeling, and the verification data is mainly used for model testing. Therefore, the main job of the model inspection phase is to use the test data into the established model, observe the response of the model, and evaluate the accuracy of the model by comparing the response of the model, thereby evaluating the accuracy of the model. If the accuracy of the model is poor, you need to re-perform data to explore, create a new model until the new model is inspected. Therefore, in practical applications, data exploration, establish model, model testing is a process of repeated iteration, see Figure 2. The process of data mining is a process of continuous exploring data characteristics, establishing and inspection models, discovering the characteristics of customer consumption behavior; mobile operators can only truly play the role of data mining by applying the results of the model to marketing and customer service.
4 Conclusion
The fierce competition of the mobile communication market has led to the construction of data warehouse for marketing and customer service and the application of data mining technology. Based on the introduction of data mining technology, the establishment of mobile communication operator data mining platform is focused on. And the theme analysis of the process and data mining based on this platform to expand data mining. It should be said that the operator's ultimate goal of various technical applications is to implement customer service-centric customer relationship management (CRM), but in the current situation, this paper believes that mobile operators are priority to establish a data warehouse On the basis of successful data mining of multi-topic.
