Benoît G, Andrews J E
College of Communication and Information Studies, School of Library and Information Science, University of Kentucky, Lexington, Kentucky, USA.
Proc AMIA Symp. 2000:61-5.
This paper is motivated by the problems of dealing with large data sets in information retrieval. The authors suggest an information retrieval framework based on mathematical principles to organize and permit end-user manipulation of a retrieval set. By adjusting through the interface the weights and types of relationships between query and set members, it is possible to expose unanticipated, novel relationships between the query/document pair. The retrieval set as a whole is parsed into discrete concept-oriented subsets (based on within-set similarity measures) and displayed on screen as interactive "graphic nodes" in an information space, distributed at first based on the vector model (similarity measure of set to query). The result is a visualized map wherein it is possible to identify main concept regions and multiple sub-regions as dimensions of the same data. Users may examine the membership within sub-regions. Based on this framework, a data visualization user interface was designed to encourage users to work with the data on multiple levels to find novel relationships between the query and retrieval set members. Space constraints prohibit addressing all aspects of this project.
本文受信息检索中处理大数据集问题的启发。作者提出了一个基于数学原理的信息检索框架，用于组织并允许终端用户对检索集进行操作。通过界面调整查询与集成员之间关系的权重和类型，有可能揭示查询/文档对之间未曾预料到的新颖关系。整个检索集被解析为离散的面向概念的子集（基于集内相似性度量），并在信息空间中作为交互式“图形节点”显示在屏幕上，最初基于向量模型（集与查询的相似性度量）进行分布。结果是一个可视化地图，在其中可以将主要概念区域和多个子区域识别为同一数据的维度。用户可以检查子区域内的成员。基于此框架，设计了一个数据可视化用户界面，以鼓励用户在多个层面处理数据，以发现查询与检索集成员之间的新颖关系。篇幅限制使得无法阐述该项目的所有方面。
