Web Personalization and Mining Using Robust Fuzzy Clustering Methods

这是一个机构间合作项目，由马里兰大学巴尔的摩县分校的阿努帕姆·乔希和科罗拉多矿业学院的拉古·克里希纳普拉姆共同开展。这项研究的目标是开发可扩展的健壮技术来对包含未知数量的重叠类别的噪声数据集进行建模，并将其应用于创建一个用于Web个性化和挖掘的软件工具。个性化有两个组成部分：(1)定制从网站提供给用户的内容；(2)探索可用的网页并对它们进行分类。该方法包括开发新的实用的聚类算法，将模糊方法、稳健统计与蒙特卡罗/自举技术相结合来对存在离群点中未知数量的重叠集进行建模。由于诸如URL、IP地址和网页的许多网络对象不能用数字特征来表示，因此正在探索新的技术来处理具有语言和文本特征的网络对象，以及通过使用这些对象之间的适当的相似性度量来对它们进行分类或聚集。正在创建和验证用于网络个性化和挖掘的软件工具，该工具将开发的算法纳入软件体系结构。该项目的结果将产生新的理论结果和有效的算法，用于同时从噪声数据集中估计未知数量的重叠类别的参数，以及将在网上提供的网络个性化和挖掘工具。因此，该项目预计将对文件的搜索和交付方式产生重大影响。它将直接影响互联网和WWW的有用性和传播性，并将从总体上促进数字图书馆技术的发展。Http://www.cecs.missouri.edu/~joshi/web-mine/

This is an inter-institutional collaborative project carried out by Anupam Joshi at the University of Maryland, Baltimore County and Raghu Krishnapuram at the Colorado School of Mines. The goal of this research is to develop scalable robust techniques to model noisy data sets containing an unknown number of overlapping categories, and apply them to create a software tool for web personalization and mining. Personalization has two components: (1) tailoring the content delivered to the user from a web site; and (2) exploring the available web pages and categorizing them. The approach consists of developing new practical clustering algorithms by combining fuzzy methods, robust statistics, with Monte Carlo/bootstrapping techniques to model an unknown number of overlapping sets in the presence outliers. Since many web objects such as URLs, IP addresses, and web pages cannot be represented by numerical features, new techniques to handle web objects with linguistic and textual features, as well as to categorize or cluster them by using suitable similarity measures between such objects, are being explored. A software tool for web personalization and mining, which incorporates the algorithms developed into a software architecture, is being created and validated. The results of this project will generate new theoretical results and efficient algorithms for simultaneously estimating the parameters of an unknown number of overlapping categories from noisy data sets, as well as a web personalization and mining tool that will be made available on the web. Thus, this project is expected to have a significant impact on the way documents are searched for and delivered. It will directly influence the usefulness and spread of the Internet and WWW, and in general, will contribute to the digital library technology. http://www.cecs.missouri.edu/~joshi/web-mine/