聚类挖掘在电信客户分类中的研究与应用

发布时间：2019-03-13 15:29

【摘要】：客户关系管理（Customer Relationship Management）简称（CRM）对改善客户与企业间的关系，有着至关重要的作用。对庞大客户信息的高效、精准的分类，是进行有效客户管理的基础和重要技术。提高电信企业竞争力的重点在于，如何将数据库中看似毫无关联不具备价值的数据通过系统的整合，提炼出有用的数据信息，并对此进行分析研究来制定差异化以及个性化的服务。数据挖掘通过数据分析，从大量的原始数据里发现其内在规律。其工作过程一是数据准备：从相关数据库中选取数据，这些数据又会被整合成可以被用于数据挖掘的数据集合；二是发现规律：使用数据准备时所得到的数据集合，采用某种方法，将这些数据集合之间的规律寻找和总结出来。本文主要针对数据挖掘技术中的一个重要领域：聚类分析进行研究分析并将其投入到现实企业决策中去。聚类挖掘其本质就是要将那些非实体的对象集合分组，使之成为由许多相似对象所形成的对象类的一个过程，其主要工作就是把收集来的那些相似的数据进行分类和聚合。本文参阅了大量的国内与国外的文献资料，对数据挖掘技术、客户关系管理系统、电信行业客户分类模型进行了深入且细致的研究。针对基于聚类挖掘的客户关系管理系统中的客户分类，开展了以下几项研究工作： 1．建立有利于数据分析的客户分类模型。客户分类模型的建立将大大提高电信企业对用户的分类，把具有某种消费习惯或者消费倾向的用户归集起来，本论文主要正对短信高频使用用户类，制定有针对性的营销策略。该客户分类模型可以将大量具有相似特征的用户归集起来，形成某个特定的客户类别，是企业能够量化的对某一用户类进行分析，制定适合的电信产品和服务。 2．从两方面入手对原有的K-means算法进行优化分析。其一本文对初始聚类中心进行优化：更好的划分效果是通过对原始聚类中心的选择而得到的。它使得不同聚类中的对象是不相似的，但一个聚类中的对象却是相似的。同时本文运用数学几何定律：“三角形两条边的长度必定大于第三条边”来降低k-means算法的总时间复杂度。以此来尽可能的达到减少迭代次数，，提高挖掘性能的目的。通过对比发现，优化后的算法比传统的K-means算法具有更好的性能。 3．确定客户分类模型中所用到的分类变量和描述变量。消费者由于各种因素之间的差异，本身也呈现出多样化，对于消费者而言，根本没有单一的策略可以对应所有客户的需求，单一的产品选择都不是一种优秀的战略选择，而客户分类在本质上就是为迅速提高一个大型多元化的组织管理水平提供了实际的可能性。客户分类以客户消费行为和客户价值为研究变量，将客户人口特征，参考客户心理消费因子作为参考依据，从而建立起了一套基于数据挖掘的客户分类模型。 4．将SPSS公司开发的Clementine7.0作为开发工具，对上述研究得出的客户分类模型通过电信客户数据进行验证，成功得出客户分类组。同时本文选取了聚类挖掘结果中的短信高频使用组为分析对象，对其消费习惯和消费倾向进行深入分析并制定出了相应的营销策略。结果表明：本文的工作将对改善电信行业的客户管理，提高用户满意度，以及提升电信运营商的市场竞争力，起到一定的支撑作用。
[Abstract]:Customer Relationship Management (CRM) plays an important role in improving the relationship between the customer and the enterprise. The efficient and accurate classification of large customer information is the foundation and important technology for effective customer management. The key point of improving the competitiveness of the telecom enterprise is how to make the data that appear to have no value in the database through the integration of the system, to extract useful data information, and to analyze the research to develop the differentiated and personalized service. Data mining, through data analysis, has found its inside gauge from a large number of raw data law. The process of work is data preparation: data is selected from the relevant database, and the data can be integrated into a collection of data that can be used for data mining; the second is the rule of discovery: the set of data obtained when using the data, and the use of a certain party The method of finding and summarizing the rules between the sets of data. This paper is mainly focused on an important field in data mining technology: the cluster analysis is used to study and analyze and put it into the decision-making of the real enterprise The essence of the clustering mining is to group the objects of the non-entities into a process of the class of objects formed by many similar objects, the main task of which is to classify and gather the similar data collected. In this paper, the author has made a deep and detailed study on the data mining technology, the customer relationship management system and the customer classification model of the telecom industry. The following research is carried out for the customer classification in the customer relationship management system based on the cluster mining. Research work:1. Establish a customer who is in favor of data analysis The establishment of the customer classification model will greatly improve the classification of the telecommunication enterprises to the users, and set the users with a certain consumption habit or the consumption tendency. The customer classification model can set a large number of users with similar features together to form a specific customer category, which can be quantified by the enterprise to analyze a certain user class and formulate a suitable telecommunication system. product and service.2. Start with the original K-means calculation from two aspects In this paper, the initial cluster center is optimized: the better partitioning effect is through the original clustering center It makes the objects in different clusters not similar, but in one cluster The object is similar. At the same time, the paper applies the law of mathematical geometry: the "The length of the two sides of the triangle must be greater than the third edge" to reduce the k-means algorithm the total time complexity can be achieved as much as possible to reduce the number of iterations, increase, The purpose of the mining performance is to compare and find that the optimized algorithm is more than the traditional K-means algorithm. with better performance.3. Determine what is used in the customer classification model Categorical variables and description variables. The difference between a variety of factors is, in itself, a variety. For consumers, there is no single policy at all to correspond to the needs of all customers, and a single product choice is not An excellent strategic choice, and the customer classification is to improve the management level of a large and diversified organization in essence The actual possibility is provided. The customer classification takes the customer's consumption behavior and the customer value as the study variable. The customer's demographic characteristics, reference customer's psychological consumption factors are used as the reference basis, and a set of data based on the data is established. and 4. using the Clementine 7.0 developed by the SPSS, as a development tool, to carry out the test on the customer classification model obtained by the above research through the telecommunication customer data; At the same time, the paper selects the short message high-frequency use group in the cluster mining result group as the analysis object, and analyzes the consumption habit and the consumption tendency. The results show that the work of this paper will improve the customer management in the telecom industry, improve the user's satisfaction, and improve the market competition of the telecom operators.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP311.13

【相似文献】