CISE Postdoctoral Research Associateships in Experimental Computer Science - An Architecture for Mining of Semi-Structured Data on the WEB

西北大学cise实验计算机科学博士后助理：网络上半结构化数据挖掘的体系结构随着网络的爆炸式增长和数字图书馆的出现，对半结构化数据的知识发现技术的需求大大增加。半结构化HTML文档的特点是数据模型丰富，即它们既包含结构化的值，也包含非结构化的概念。探索这些非结构化概念之间关系的语义知识是重要的，但作为一个在线过程，成本很高。博士后助理将协助开发一个原型Web挖掘系统，该系统将允许其用户指定一组约束，数据值或概念，并获得一组包含给定约束和与约束集相关的其他项的规则。为了导出这些规则，系统使用了一个概念库来区分一般概念和与特定领域相关的概念。具体来说，该协会将1)开发新方法来生成部分约束关联规则和分类规则，这些规则使用库中保存的扩展概念层次结构；2)研究各种可用的词汇工具，并查看如何将它们集成到Wordnet中（这构成了概念库的基础）；3)探索各种启发式方法，以确定给定的概念是属于专门的概念库还是属于通用的概念库。4)探索使用采样技术减少规则生成的搜索空间。

EIA-0000536Scheuermann, PeterNorthwestern UniversityCISE Postdoctoral Associates in Experimental Computer Science: AnArchitecture for Mining of Semi-Structured Data on the WebWith the explosion of the Web and the advent of digital libraries, thedemand for knowledge discovery techniques for semi-structured data hasincreased greatly. Semi-structured HTML documents are characterized by arich data model, i.e., they contain both structured values and unstructuredconcepts. Exploring the semantic knowledge about the relationships amongthese unstructured concepts is important but costly as an on-line process.The postdoctoral associate will assist in developing a prototype Web miningsystem that will allow its users to specify a set of constraints, eitherdata values or concepts, and obtain a set of rules that contain the givenconstraints and other items related to the constraint set. In order toderive these rules, the system makes use of a concept library thatdifferentiates between general concepts and concepts associated withspecialized domains. Specifically, the associate will 1) develop newmethods for generating partially constrained association rules andclassifications rules that make use of the extended concept hierarchieskept in the library, 2) investigate the various lexical tools available andsee how to integrate them into Wordnet (which forms the basis of theconcept library), 3) explore various heuristic methods for deciding when agiven concept belongs to a specialized concept library or to the generalconcept library, and 4) explore the use of sampling techniques in reducingthe search space of rule generation.