Knowledge Discovery and Mandgement for Information Network

在科学和商业应用中，例如信息系统、通信网络、生产过程等，大量的数据被快速、连续地生成，并存储到管理数据库中。为了对这样的系统进行控制，建立数学模型进行分析，并从规律数据中发现有用的知识是非常重要的，本课题主要研究了以下几点：1.数学模型的混合性能评价系统提出了一种基于淬火理论的模型描述语言。我们使用面向对象语言开发了一个集成软件环境的原型。基于我们开发的知识库，我们的软件环境可以提供非常可靠的，灵活的和用户友好的手段，性能评估的目的。分析器和模拟器在分布式环境下协同工作。2.信息网络中的知识发现我们可以从信息网络资源中发现原始数据的函数或多值依赖规则或完整性约束的形式。我们扩展了面向属性的算法，在底层处理数据和高层可理解的控制原语之间架起了桥梁。通过该归纳算法发现的知识也可以与统计信息相关联。我们提出的算法发现三种有趣的规则：特征规则，稳定规则和进化规则。主动数据库技术提供了快速、实时、智能化的管理。它可以大大提高信息网络中数据挖掘的能力。

In scientific and business applications, such as a information system, a communication network, a production process, etc., huge volume of data is generated rapidly and continuously, and stored into management database. In order to control such a synamic system, it is important to make mathematical model for analysis and to discover the useful knowledge from the law data.Our project focused on the following points.1.Hybrid Performance Evaluation System for Mathematical ModelsWe propose a model description language which is based on queneing theory. We developed a prototype of integrated software environment using object oriented language. Based on our developing konwledge base, our software environment can provide very reliable, flexible and user-friendly means for the performance evaluation purpose. Analyzer and simulator work in cooperation with each other on distributed environment.2.Knowledge Discovery from Infomation NetworkWe can discover the form of functional or multi-valued dependency rules or integrity constraints in terms of primitive data from the resource of information network. We extended the attribute-oriented algorithm to make the bridge between the low-level processing data and high level understandable control primitives. Discovered knowledge by this induction algorithm can be also associated with statistical information. Our proposed algorithm discovers three kinds of interesting rules : characteristic rule, stable rule and evolution rule. The technology of active database offers the prompt, real-time, and intelligent management. It could substantially enhance the power of data mining in information network.