【作者】 张捷；

【导师】 琚春华；

【作者基本信息】 浙江工商大学 ， 管理科学与工程， 2007， 硕士

【摘要】 面对日趋激烈的商业竞争，各企业纷纷走上信息化道路，通过ERP提高企业的商业竞争力。进而，对这些海量的ERP数据进行数据挖掘，从中得到潜在的、有用的知识，以辅助商业决策。但是随着企业的不断发展壮大，原来在企业中起过很大作用的数据库管理系统的数据量成倍增加，其管理逐步由集中式向分布式发展，如何有效地对这些分布的数据库进行数据挖掘成为新的挑战。传统的数据挖掘基本上是一个本地的数据分析工具，仅能对本地数据集产生一定的理解性或概括性的知识，而在数据分布环境下，除了结点上是物理分布的，处理的是海量数据，同时还要兼顾数据的安全性以及非共享数据的隐私性。针对这些问题，以数据挖掘中的分类和预测为重点，本文提出了基于贝叶斯网络的分布式商务数据挖掘模型(DDMMBN，Distributed Data Mining Model Based on BayesianNetwork)，该模型是以具有移动Agent功能的Bee-gent系统为框架，以贝叶斯网络相关性学习理论为方法，以属性多叉树为中间过程，从分布的商业数据库中训练得到综合的贝叶斯网络，利用综合的贝叶斯网络推理实现对客户的分类和消费量的预测。该模型(DDMMBN)中提出了属性多叉树这一数据结构，该属性多叉树能反应各分布的数据集的属性特征值，它可以通过移动Agent访问各分布的数据集，调用其属性多叉树构建算法而得到，然后利用属性多叉树得到贝叶斯网络。该属性多叉树能很好地解决数据分布的问题，不需要将各分布的数据汇总，大大地减轻了网络负担，节省了本地存储空间。同时，由于该属性多叉树只是概括了分布数据集的特征值，而不需要涉及每条数据记录的细节，故在一定程度上能很好地解决其分布数据的隐私性问题。本文在详细阐述了贝叶斯网络理论、分布式数据挖掘相关技术和移动Agent技术后，针对商业企业中客户的分类和消费量的预测问题，提出了该基于贝叶斯网络的分布式商务数据挖掘模型(DDMMBN)。以Bee-gent系统为基础，建立了该模型的原型系统，利用已有的商业数据，与数据汇总法和加权表决法相比较，证明了其具有较高的时间效率和分类准确率。更多还原

【Abstract】 Facing with increasingly fierce competition, information system has been used extensively in the enterprises, the commercial competitiveness of which could been enhanced by ERP. In turn, we can get some potential and useful knowledge from mining the massive data of ERP, which could be used to support business decisions. However, as the business develops and grows, the volume of data of database management which originally played a significant role increased greatly in the enterprise, so the management way of database changed gradually from centralization to distribution. How to mining these distributed database became a new challenge.The traditional data mining was just a local data analysis tool, which only can get some understandable or general knowledge from local datasets, but in the distibuted datasets environment, the node was physical distributed, and the processing data was massive, and the security and privacy of not-sharing data must be considered. For these problems, focusing the classification and prediction of data mining, a distributed data mining model based on bayesian network(DDMMBN) was introduced in this paper. This model used the Bee-gent system which has the function of mobile agent as the framework, and used the relational learning of bayesian network as the way, and used the multi-branches tree of attribute as the middle process, got the integrated bayesian network from distributed business database by learning, and use this bayesian network inference to realize the classification of customers and forecast of consumption.The multi-branches tree of attribute was intruduced in this model, which could reflect attribution eigenvalues of distributed datasets. It can be gotten by the mobile agent which accessed the distributed datasets and call algorithm of building the multi-branches tree of attribute, and then the multi-branches tree of attribute was used to creat a bayesian network. The multi-branches tree of attribute could solve the distributed problem well, it didn’t need to collect all the data, which greatly reduced the burden on the network and save the local storage space. Meanwhile, because the multi-branches tree of attribute just only include the eigenvalues of attribute, which not involve the details of each data record, to a certain extent it can be a very well solution for the privacy issues of distribution datasets.In this paper, after particular explanation of Bayesian network theory, distributed data mining technology and mobile agent technology, for the customer classification and consumption forecasts in the commercial enterprises. a distributed data mining model based on bayesian network(DDMMBN) was introduced. Based on the Bee-gent system, we have built a prototype system, and used existing business data, compared with the data collection method and the weighted voting method, prove its efficiency and high classification accuracy.更多还原