Berrios D C
Veterans Affairs Palo Alto Health Care System, Palo Alto, CA, USA.
Proc AMIA Symp. 2000:71-5.
We report our experience with a statistically based method of generating sentence-level indexing based on identified UMLS concepts and query and vector-space models. We evaluated the system using the consensus markup of two domain experts as the gold standard. UMLS concepts identified both from HTML headings and in paragraph text were valuable in proposing markup. Using both sources of concepts, the model proposed the correct set of concepts in the form of a query prototype 71% of the time. The correct query prototype was ranked first or second in 79% of cases.
我们报告了我们基于统计方法生成句子级索引的经验,该方法基于已识别的统一医学语言系统(UMLS)概念以及查询和向量空间模型。我们使用两位领域专家的一致标注作为金标准来评估该系统。从HTML标题和段落文本中识别出的UMLS概念在提出标注方面很有价值。使用这两种概念来源,该模型在71%的情况下以查询原型的形式提出了正确的概念集。在79%的案例中,正确的查询原型排名第一或第二。