基于三元组特征和词向量技术的中文专利侵权检测研究  

Infringement detection of Chinese patent based on three tuple character and word embedding

作　　者：金健[1] 朱玉全[1] 陈耿[2] 

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013 [2]南京审计大学工学院,南京211815

出　　处：《计算机应用研究》2017年第10期2901-2904,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(71271117);江苏省六大人才高峰项目(2013-WLW-005);江苏省自然科学基金资助项目(BK20150531)

摘　　要：针对中文专利侵权检测中关键词特征表达能力弱以及句子结构特征容易引起噪声干扰的问题,提出了一种通过抽取三元组特征来改进中文专利侵权检测的方法。该方法将专利权利要求书抽取为三元组特征的集合,并结合词向量技术和How Net计算三元组特征间的语义相似度,从而有效提高对疑似侵权专利的识别能力。实验结果表明,该方法取得了较好的检测效果,且在准确率上要高于其他方法。Because the expression ability of keywords features are weak and the structural features of the sentence are easy to cause the problem of noise interference, this paper proposed a method of improving Chinese patent infringement detection by extracting the three tuple features of the claim. This method extracted the patent claim as a set of three tuple, and calculated the similarity between the three tuple features by combining word embedding and HowNet, which could effectively improve the ability to identity the suspected patent infringement. Experimental results show that the proposed method has good detection resuits, and the accuracy is higher than other methods.

关 键 词：专利侵权 信息抽取 词向量 相似度计算 文本处理 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]