Using citation data to improve retrieval from MEDLINE

描述（由申请人提供）： 信息过载不再是一个理论概念，而是教育、研究和患者护理的真实的障碍。我们的长期目标是通过为学生，研究人员和临床医生提供更好的信息检索工具来改善患者护理。我们的统一假设是，万维网（WWW）上开创的技术可以成功地适应MEDLINE和科学引文索引（SCI）的组合，以提高信息检索。MEDLINE和SCI的结合是一个类似于WWW的超链接环境。因此，成功的WWW算法可以应用于识别最重要的和相关的文章，以满足用户的信息需求。本提案的具体目的是：1）评价引文分析对MEDLINE检索结果排序的益处; 2）评价引文分析对确定MEDLINE中文章相似性的益处。这项研究是在斯坦福大学的NLM奖学金培训期间进行的工作的延续，我们在那里设计和实施了Medline查询的例子（MQBE）计算框架。MQBE框架将用于最大限度地减少执行和评估信息检索策略所需的人力。计算框架将被增强以存储信息需求的陈述、查询和正在使用系统来满足真实的信息需求的真实的用户的相关性判断。将评价结果排序算法优先返回专家小组选定的“关键文章”的能力。此外，Joachims开发的一种新的评估方法，使用点击率来比较替代搜索策略，将被用来比较算法。本提案中收集的数据将作为可持续的一般研究工作的基础，重点是信息检索。

DESCRIPTION (provided by applicant): Information overload is no longer a theoretical concept, but a real impediment to education, research and patient care. Our long-term goal is to improve patient care by providing better information retrieval tools to students, researchers and clinicians. Our unifying hypothesis is that techniques pioneered on the World Wide Web (WWW) can be successfully adapted to the combination of MEDLINE and the Science Citation Index (SCI) to improve information retrieval. The combination of MEDLINE and SCI is a hyper linked environment similar to the WWW. Therefore, successful WWW algorithms can be applied to identify the most important and relevant articles to fulfill users' information needs. The specific aims of this proposal are: 1) To evaluate the benefit of citation analysis for ranking MEDLINE search results and 2) To evaluate the benefit of citation analysis for determining article similarity in MEDLINE. This research is a continuation of work performed during NLM fellowship training at Stanford where we designed and implemented the Medline Query-by-Example (MQBE) computational framework. MQBE framework will be used to minimize the human effort required to implement and evaluate information retrieval strategies. The computational framework will be enhanced to store statements of information need, queries and relevance judgments of real users who are using the system to fulfill real information needs. Result ranking algorithms will be evaluated with respect to their ability to preferentially return "key articles" selected by panels of experts. In addition, a new evaluation methodology developed by Joachims that uses click throughs to compare alternative search strategies will be employed to compare algorithms to each other. The data collected in this proposal will serve as the foundation for a sustainable general research effort focusing on information retrieval.