Advanced models and Techniques for Expert Interactive Retrieval Assistance

专家交互式检索协助的高级模型和技术

• 批准号: 8805260
• 负责人: Richard Marcus
• 金额: $ 11.89万
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1988
• 资助国家: 美国
• 起止时间: 1988-09-01 至 1991-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8805260&HistoricalAwards=false
• 关键词: Advanced; models; Techniques; Expert; Interactive

Enhancement, experimental analysis and evalution will be accomplished on an expert retrieval assistant named CONIT which seeks to capture and formalize some of the intelligence embodied in a human expert searcher. In particular CONIT assists users in developing a formalized problem description from which the system derives and executes a search strategy in one or more of 400 databases in three different retrieval systems. CONIT further assists the user in displaying document records and reformulating the search strategy based on search results and user desires. New assistance techniques to be studied include (1) ranking documents by estimated relevance based on retrieval match criteria; (2) several means for search evaluation based on estimates of recall plus time and cost parameters; and (3) selection of the most propitious search modification tactics for the current problem context. With these tools search strategy planning can be based on more rational, quantitative decision- making procedures as opposed to the more intuitive, qualitative procedures typical of current retrieval modalities. Analysis of experiments conducted will consider how well the particular techniques, and general retrieval models (subsumed under the term "smart Boolean"), achieve these goals. An improved human- computer interface will also be tested through conversion of the CONIT system from a mainframe to a workstation environment.

增强、实验分析和评估将在一个名为CONIT的专家检索助手上完成，该助手旨在捕获和形式化人类专家搜索器中体现的一些智能。特别是，CONIT帮助用户开发形式化的问题描述，系统从中派生并在三个不同检索系统的400个数据库中的一个或多个数据库中执行搜索策略。CONIT进一步帮助用户显示文档记录，并根据搜索结果和用户需求重新制定搜索策略。需要研究的新辅助技术包括：(1)基于检索匹配标准的估计相关性对文档进行排序；(2)几种基于召回率加时间和成本参数估计的搜索评估方法；(3)针对当前问题情境选择最有利的搜索修改策略。有了这些工具，搜索策略规划可以基于更理性、定量的决策程序，而不是基于当前检索模式中更直观、定性的典型程序。所进行的实验分析将考虑特定技术和一般检索模型（包含在术语“智能布尔”下）如何实现这些目标。通过将CONIT系统从大型机转换到工作站环境，还将测试改进的人机界面。