33 疾病名词识别模型的训练.doc

针对海量新闻的弱监督式疾病名词识别 陈瑛a* 裴蓓b* 彭珺a 程碧霄a 赵明a 杜亚茹a a 中国农业大学 E-mail: chenying@ b 信息网络安全公安部重点实验室 E-mail: peibei@ 摘 要：针对新闻报道的敏感性内容，疾病相关报道是新闻审核部门的一项重要审查内容，本文研究一种基于海量新闻的疾病名词自动识别方法，其能基于现有的疾病领域本体，综合利用网络爬虫和信息抽取技术，最终完成对海量新闻报道进行自动标注。首先，虽然表明 关键词：疾病；专有名词识别；支持向量机。 Weakly-supervised Disease Names Recognition for a Large-scale News Reports Abstract: As a kind of sensitive information in content of news, disease-related reporting is an important auditing content of news audit department. Based on this background, this paper studied a method of automatic recognition of disease names in Chinese mass news reports. Based on disease ontology, we use of web crawler and information extraction comprehensively to tag the news reports automatically. Firstly, Traditional support vector machine approach is based on a supervised Machine Learning, whose recognition effect is relatively good, but it relies on high quality of large-scale artificial tagged corpus. Therefore, this paper proposes a weakly supervised method of automatic recognition of disease names based on ontology learning and SVM method. Secondly, taking into account most of the existing disease domain ontology use of academic terminology, but the news report is for the people. The language is relatively popular, so this paper expands the alias of disease terms. Experiments show that this study can automatically identify the disease specific name, providing a reference for the identification and analysis of mass media news sensitive content. Keywords: disease, name-entity recognition, Support Vector Machine。 1 引言一个重要的研究方向 因为疾病名词属于专有名词，而专有名词抽取技术是信息抽取的重点研究内容，所以该识别任务其实隶属于信息抽取研究范畴, 其主要任务是专有名词识别和关系识别[1]。到目前为止，专有名词识别方法主要采用机器学习方式，其利用人工标注语料作为训练语料，结合各种全监督式机器学习方法和特征提取方法，得到专有名词识别模型。由于，人工标注语料费时费力，并且面临覆盖率差等各种问题，因此需要研究高质量的训练语料自动标注方法。 本体作为一种能够在语义和知识层面上描述事物的概念模型，可以提供某一领域的术语及其关系，从而可以用于语料的自动标注。目前国际和国内已经存在一些通用的疾病分类体系，这些体系一般采用相对规范的学术用语，而 针对以上这些问题，本文提出一个基于支持向量机的弱监督式疾病专有名词自动识别方法。 2 相关研究 到目前为止，专有名词识别方法主要有以下几类：基于规则，基于统计，以及二者的相结合。 基于规则的专有名词识别大多利用人工定制的规则提