Heuristics for identification of acronym-definition patterns within text: Towards an automated construction of comprehensive acronym-definition dictionaries
Heuristics for identification of acronym-definition patterns within text: Towards an automated construction of comprehensive acronym-definition dictionaries
复制标题
DOI：
10.1055/s-0038-1634373
复制
发表时间：
2002-01-01
影响因子：
1.7
通讯作者：
Garner, HR
中科院分区：
文献类型：
--
作者：
Wren, JD;Garner, HR
目的：开发一种自动化、准确和可扩展的方法，通过该方法可以在文本中识别首字母缩略词-定义对。它的主要优势是使信息处理方法能够解析作者定义的缩略语，但它还允许自动创建关于缩略语定义的参考工作。除了节省时间和精力外，这与人工或半自动方法相比有几个优点，例如能够识别备选缩略语和定义的相对频率，以及唯一的缩略语-定义对的拼写、短语和连字变体。它还帮助用户识别文献中可能不一定存在于生物医学数据库中的缩略语/定义变体。方法：从MEDLINE记录子集的精确度和召回率方面开发并改进了一套启发式方法，以准确定位和识别缩略语-定义对的边界。结果：我们的最终缩略语解析通用启发式算法(ARGH)基于样本的估计准确率为96.5+/-0.4%，召回率为93.0+/-2.7%，对超过1200万条MEDLINE记录进行测试，识别了超过174,000个唯一的缩略语及其737,000个相关定义。结论：我们估计MEDLINE中36%的缩略语与一个以上的定义相关，相反，高达10%的定义与一个以上的缩略语相关。MEDLINE中独特的缩略语的数量正在以每年约11,000个的速度增长，而与它们相关的定义的数量以大约四倍的速度增长。可通过http://lethargy.swmed.edu/ARGH/argh.asp.在线访问ARGH数据库启发式模块和数据库可根据要求提供。
Objectives: To develop an automated, accurate and scalable method by which acronym-definition pairs can be identified within text. Its primary advantage is in enabling information processing methods to resolve author-defined acronyms, but it also allows an automated creation of a reference work on acronym definitions. This has several advantage over manual or semi-automated methods, besides time and effort saved, such as enabling identification of relative frequencies for alternate acronyms and definitions as well as spelling, phrasing and hyphenation variants for a unique acronym-definition pair. It also aids users in identifying acronym/definition variants present in the literature that may not necessarily be in biomedical databases.Methods: A set of heuristics to accurately locate and identify the boundaries of acronym-definition pairs was developed and refined in terms of precision and recall on subsets of MEDLINE records. These training sets were gradually increased in size and heuristics re-evaluated to ensure scalability.Results: Our final set of Acronym Resolving General Heuristics (ARGH) had a sample-based estimated rate of 96.5+/-0.4% precision and 93.0+/-2.7% recall when tested on over 12 million MEDLINE records, identifying more than 174,000 unique acronyms and their 737,000 associated definitions.Conclusions: We estimate that as much as 36% of the acronyms in MEDLINE are associated with more than one definition and conversely up to 10% of definitions are associated with more than one acronym. The number of unique acronyms in MEDLINE is increasing at a rate of approximately 11,000 per year, while the number of definitions associated with them is growing at approximately four times that rate. Access to the ARGH database is available online at http://lethargy.swmed.edu/ARGH/argh.asp. The heuristic module and database are available upon request.