一种基于自举原理的语义模式自动获取方法  

A Bootstrapping Method for Automatic SemanticPattern Acquisition

作　　者：李向阳[1] 张亚非[2] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007 [2]解放军理工大学训练部,江苏南京210007

出　　处：《微电子学与计算机》2005年第2期188-192,共5页Microelectronics & Computer

基　　金：国家自然科学基金项目资助:(60303024)

摘　　要：语义模式的自动获取是解决信息抽取系统可移植性问题的关键之一。文章提出了一种基于自举原理的语义模式自动获取方法。该方法是一种新的双循环自举机制,由内自举和外自举两部分组成。内自举借助通用的《同义词词林》扩张模式的语义范围;外自举利用模式之间的交叉项作为链接点自动寻找新模式。用于从未标注的文本中自动发现语义模式,启动时只须用户提供少量的抽取示例作为种子,运行中即可增量式地发现新语义模式。测试表明,该方法产生的语义模式在信息抽取中具有较好的准确率和召回率。Automatic semantic pattern acquisition is a key aspect for addressing the portability problem of information extraction systems. A method based on bootstrapping principle is presented to automatically acquire semantic patterns. The method has a novel dual-cycle bootstrapping mechanism composed of an inner bootstrapping and an outer bootstrapping. The inner bootstrapping utilizes Synonymy Thesaurus to expand pattern semantics. The outer bootstrapping uses intersections among patterns as link points to find new patterns. Starting with a small set of information extraction samples as seeds, the method can incrementally identify new patterns from un-annotated text. Tests show semantic patterns produced with the method have good result for both precision and recall in information extraction.

关 键 词：语义模式 自举法 信息抽取 模式获取 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]