基于CRFs的冶金领域中文专利术语抽取研究

发布时间：2019-03-27 11:17

【摘要】：【目的】探讨冶金领域中文专利术语抽取模型的最优条件,用于有效地抽取冶金领域专利术语。【方法】使用尚不完善的核心语料库,在无需人工标引的情况下,采用条件随机场(CRFs)构建字角色标注的冶金领域中文专利术语识别模型。详细说明模型的构建过程,同时重点对比CFRs的各个因素(特征组合、字长窗口等)对识别效果的影响。【结果】实验结果表明字序列、级别特征、领域特征、温度特征的组合在字长窗口为3,c等于1,f等于1时,准确率达到94.26%,召回率达到94.37%,F1值达到94.5%。【局限】核心词典欠完善,使得部分词语标注不够准确;未与其他方法作详细比较,未详细说明CRFs的可靠性。【结论】CRFs在适当的角色和特征以及特征模板的组合下能较好地识别出冶金领域的中文专利术语。
[Abstract]:[objective] to explore the optimal condition of Chinese patent term extraction model in metallurgical field, which can be used to effectively extract patent term in metallurgical field. [methods] the core corpus, which is not perfect, can be used without manual indexing. The conditional random field (CRFs) is used to construct the Chinese patent term recognition model of character tagging in metallurgical field. The construction process of the model is described in detail, and the influence of each factor of CFRs (feature combination, word length window, etc.) on the recognition effect is compared in detail. [results] the experimental results show that the word sequence, the level feature, the domain feature, and so on. The combination of temperature features at the word length window of 3, c equals 1, f equals 1, the accuracy rate reaches 94.26%, the recall rate reaches 94.37%, and the F1 value reaches 94.5%. Make part of the words not accurate tagging; The reliability of CRFs is not explained in detail compared with other methods. [conclusion] CRFs can identify the Chinese patent terms in metallurgical field well under the combination of appropriate roles and features and feature templates.
【作者单位】：南京大学信息管理学院;江苏省数据工程与知识服务重点实验室;
【基金】：江苏省自然科学基金项目“面向专利预警的中文本体学习研究”(项目编号:BK20130587) 江苏省“333”工程项目“面向知识服务的中文本体学习研究”(项目编号:BRA2015401)的研究成果之一
【分类号】：TP391.1;G306

【参考文献】