| 基于图卷积网络的高质量专利自动识别方案研究 |
| |
| 引用本文： | 吴洁,桂亮,刘鹏.基于图卷积网络的高质量专利自动识别方案研究[J].情报杂志,2022(1). |
| |
| 作者姓名： | 吴洁 桂亮 刘鹏 |
| |
| 作者单位： | 江苏科技大学经济管理学院 |
| |
| 基金项目： | 国家社会科学基金后期资助项目“产学研联盟主体知识转移博弈和创新绩效研究”(编号:19FGLB029);国家自然科学基金面上项目“多层复杂网络视角下开源软件社区集体智慧涌现机制研究”(编号:71871108);江苏高校哲学社会科学研究重点项目“资源双依赖视角下后发企业创新机理与实现研究”(编号:2018SJZDI053);工信部2020年先进制造业集群项目“江苏省通泰扬海工装备和高技术船舶集群”(编号:TC200J023)研究成果之一。 |
| |
| 摘 要： | 研究目的]高质量专利对促进专利转化、技术追踪和战略布局十分重要,面对海量专利数据,如何准确高效自动识别高质量专利,为开展后续专利投资融资、产业转型等专利工作做基础铺垫,成为当前重要研究问题。研究方法]以国家知识产权局受理的申请专利为研究对象,使用专利维持年限表征专利质量,提取专利数字特征并嵌入专利文本特征生成的专利-核心词汇网络,搭建图卷积网络模型自动识别高质量专利。研究结论]目前针对专利质量的研究专注于挖掘专利数字特征而忽视专利文本特征,该方案在高质量专利自动识别过程中使用专利数字特征与文本特征,对当前专利质量研究做出补充。此外,所提方案可在专家标注少量专利文档情况下完成专利质量识别任务,解决现有专利质量标签标注方案无法全面衡量专利质量的局限。同时,将图卷积网络扩展到专利背景下的质量识别领域,为专利质量研究提供崭新框架,实验结果也显示方案具有较高实践价值。
|
| 关 键 词： | 专利质量 图卷积网络 评价指标 文本特征 自动识别 |
Indicator and Textual Features-Based Patent Evaluation with Graph Convolutional Networks |
| |
| Wu Jie,Gui Liang,Liu Peng.Indicator and Textual Features-Based Patent Evaluation with Graph Convolutional Networks[J].Journal of Information,2022(1). |
| |
| Authors: | Wu Jie Gui Liang Liu Peng |
| |
| Institution: | (School of Economics and Management, Jiangsu University of Science and Technology, Zhenjiang 212003) |
| |
| Abstract: | Research purpose]High quality patents are extraordinary significant to promote the patent transformation,technological tracking and strategic layout.Faced with massive patent data,how to accurately,efficiently and automatically identify high quality patents has become an crucial research issue at present.Research method]Therefore,the patent application accepted by CNIPA(China National Intellectual Property Administration)is taken as the research object.And this study uses patent life indicator to determine the type of patent quality,chooses patent indicators related to patent quality as patent number features,then patent number features are embedded into the patent-vocabulary network which was generated by patent text features for generating high quality patent automatic evaluation model based on Graph Convolutional Network.Research conclusion]Current research on patent quality focuses on mining patent number features while ignoring patent text features.In this study,patent number features and patent text features are used in the automatic recognition process of high-quality patents to supplement the current research on patent quality.In addition,the approach proposed in this paper can complete the task of patent quality identification under the condition that experts mark a small number of quality labels of patent documents,which solves the limitation that the patent indicator labeled patent document quality scheme cannot fully measure the patent quality.Moreover,the graph convolutional network is extended to the field of patent quality automatic evaluation,which provides a new framework for patent quality research.Experimental results also show that the proposed approach performs better practical value. |
| |
| Keywords: | patent quality graph convolutional network evaluation indicator textual feature automatic evaluation |
| 本文献已被 维普 等数据库收录！ |
|
