Knowledge Discovery in Temporal Databases

从金融市场分析到天气预报，从监控超市采购到监控卫星图像，越来越多的应用领域都变得越来越数据密集。因此，数据库中的知识发现领域近年来引起了数据库研究者的极大兴趣。由于许多数据挖掘应用程序本质上是时态的，因此研究时态数据库上下文中的模式发现问题非常重要。这个项目的目的是：(1)为时间模式的特征和分类建立一个框架；(2)在提出的框架中识别和描述各种类型的模式发现问题；(3)评估人工智能、运筹学、信号处理、统计时间序列分析和其他相关学科的现有技术对第(2)部分问题的适用性，并在必要时开发新的时间模式发现技术——如果可能的话，这些技术应该更有效；(4)并行化发现算法，使其在计算上可行；这一点很重要，因为时间模式发现问题通常要处理大量数据。作为这个项目的结果，将开发一个系统，帮助用户从大量的时间数据中发现知识。本文的研究将对时态数据库中的模式发现问题提供更好的理论理解，并为时态数据中的模式发现提供一些实用的工具。这项工作的潜在应用包括金融、营销和医疗等方面的应用。＊＊＊

*** 9318773 Tuzhilin More and more application domains, from financial market analysis to weather prediction, from monitoring supermarket purchases to monitoring satellite images, are becoming increasingly data-intensive. For this reason, the area of knowledge discovery in databases has recently attracted much interest of database researchers. Since many data mining applications are temporal in nature, it is important to study the problems of pattern discovery in the temporal database context. The purpose of this project is to (1) develop a framework for the characterization and classification of temporal patterns; (2) identify and characterize various types of pattern discovery problems within the proposed framework; (3) evaluate the applicability of existing techniques from artificial intelligence, operations research, signal processing, statistical time-series analysis, and other related disciplines to the problems from part (2), and to develop new temporal pattern discovery techniques, whenever necessary -- also these techniques should be made more efficient, if possible; and (4) parallelize the discovery algorithms in order to make them computationally feasible; this is important because temporal pattern discovery problems typically deal with large volumes of data. As a result of this project, a system will be developed that helps a user to discover knowledge from a large volume of temporal data. This research will provide a better theoretical understanding of the problems of pattern discovery in temporal databases, as well as provide some practical tools for finding patterns in temporal data. Potential applications of this work include financial, marketing, and medical applications, among others. ***