【作者】 宋艳；

【导师】 梁静国；

【作者基本信息】 哈尔滨工程大学 ， 管理科学与工程， 2004， 博士

【摘要】 在竞争激烈的商业时代，资源占有成为决定企业生死成败的关键。在客户关系方面，企业总希望建立与客户最稳固的关系，并最有效率地把这种关系转化为利润，即留住老客户、发展新客户并锁定利润率最高的客户，这也就是CRM要重点研究的问题。为了实现这个目标，企业就需要尽可能地了解客户的行为，但这种了解不可能通过与客户接触直接获得，因为企业不可能挨个与客户交谈，而且他们所需要的信息，单个客户往往无法提供。企业所能做的，就是尽可能收集客户的信息，借助各种分析方法，透过无序的、表层的信息挖出内在的知识和规律，即利用数据挖掘技术实现CRM管理理念。在挖出大量信息之后，企业就可以根据这些规律或用这些信息设计数学模型，对未发生行为做出结果预测，为企业的综合经营决策、市场策划提供依据，从而体现CRM管理理念。 在CRM应用系统中，采用嵌入数据挖掘系统的方式，利用数据挖掘技术从大量的有关客户的数据中挖掘出隐含的、先前未知的、对企业决策有潜在价值的知识和规则。本文通过对客户关系管理和聚类数据挖掘技术研究，提出基于改进的CABOSFV算法的客户聚类算法，用于解决客户关系管理中，对由大量高维稀疏数据组成的客户行为数据集进行聚类分析。 由于一个数据库或者数据仓库都有很多的维，一些分析算法在处理维数比较少的数据集时表现不错，例如两、三维的数据；人的理解能力也可以对两、三维数据的聚类分析结果的质量作出较好的判别，但对于高维数据就没有那么直观了。所以对于高维数据的聚类分析是很具有挑战性的，特别是考虑到在高维空间中，数据的分布是极其稀疏的，而且形状也可能是极其不规则的。所以，本文针对高属性维稀疏数据的聚类数据挖掘技术进行了研究。首先，研究高属性维稀疏聚类的算法，提出集合差异度的定义方式、集合差异度阈值的计算公式，从而改进CABOSFV算法；然后，在研究大数据集对象的数据消减策略的基础上，阐述如何采用采样策略进行数据消减，并利用集合哈尔滨工程大学博士学位论文石面‘亩亩‘面面面面面面函面面面上下确界的概念，完成非样本对象向基础类的匹配;其次，为进一步完善聚类的应用，还进行了异常数据的挖掘研究，并为弥补采样策略给聚类带来的概率缺陷，提出了孤立点对象的处理方案;最后，针对某药厂客户购买行为，提出基于客户购买能力的稀疏特征的转换算法，重新构造了聚类过程，并以该药厂的药品销售记录为聚类数据源，进行了聚类过程的实证检验及分析。关键词:聚类;数据挖掘;客户关系管理;CABOSFV算法更多还原

【Abstract】 In the commercial epochs where there is keen competition, possessing resource is fatal to decide enterprises livings. For customer relationship, enterprises always wish to build the most steady relationship and transfer it into profit effectively. It also can be said that keep formal customers, developing new customers and locking those very important customers. This is what CRM want to study mainly. For realizing this goal, enterprises need do their best to know customers behaviors. However, they can’ t meet with customers one by one to get information. What they can do is that they do their best to collect information and then discover the inner knowledge and rules from disordered and surface information by kinds of analysis methods, which also can be named Data Ming. After mining mass information, enterprises can design mathematical models according to these rules or information and forecast the unknown results to provide basis for selling decision and market scheme to exhibit CRM.In CRM applied system, data mining system is embedded in and data mining is used to seek implications, unknown before and valuable knowledge and rules benefit to enterprises decision from mass data related to customers. This theme put forward to a kind of customers clustering algorithm based on improving CABOSFV algorithm to solve how to cluster mass high-dimension square data which express customers behaviors after studying CRM and clustering Data Mining technology. There are many dimensions in a database or data warehouse. Some algorithms are good for dealing with few dimensions such as two-dimension or three-dimension data. People can distinguish the qualities of few-dimension data clustering results easily, but the results of many-dimension are not so visual. So, clustering analysisof high-dimension data is challenging. Especially, the distribution of data in high-dimension is very square and not irregular. Thus, data mining technology of clustering high-dimension square data is studied. First, the algorithm of high-dimension square data is mainly studied. Putting up definition of set-difference-degree and calculating formula of threshold to improve the algorithm of CABOSFV. Second, because clustering objects are usually in mass data set, data-reduce-strategy in mass set is studied. How to use sample to reduce data is stated and match non-sample to basis clusters using the conceptions of set upper and lower bound. Third, for perfecting clustering utilities more, mining anomalous data is studied and solution of isolated objects is put forward to make up the probabilities-fault of samples-cluster. Last, the boundary of data mining technology based on clustering analysis using in analytical CRM applied system is studied and transferring algorithm of square feature based on customers purchasing abilities is put forward, then rebuild clustering procedure. Selling medicine of some pharmaceutical factory is used to check and analyze the new procedure.更多还原