基于机器学习的专利质量评估研究  

Estimation of Patent Quality Based on Machine Learning

作　　者：杨美妮[1] 沈静[1] 张建军[1] YANG Meini;SHEN Jing;ZHANG Jianjun(College of Science, Naval University of Engineering, Wuhan 430033)

机构地区：[1]海军工程大学理学院

出　　处：《计算机与数字工程》2019年第7期1743-1747,共5页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61402516);海军工程大学自主立项项目(编号:2016108)资助

摘　　要：专利中含有丰富的计量信息,这些计量信息与专利质量之间存在着一定的联系。论文利用3D打印相关的六千多个专利族的十一组专利计量信息,测试了逻辑回归、支撑向量机和神经网络三种机器学习模型在专利质量评估上的表现。神经网络模型获得了较好的实验效果,运用逻辑回归模型可以对准确率或召回率进行单边控制,从而满足特定的应用需要。Patent contains rich metric information which is related to patent quality.In this study,the information of patent is used to estimate its quality based on three machine learning models.The models include logistic regression,support vector machine and neural network.More than 6,000 patent families in 3D printing field have been evaluated by these three methods.This study shows that neural network achieves the best result,and the logistic regression can be used in precision and recall control to meet the needs of specific applications.

关 键 词：机器学习 专利质量 逻辑回归 支撑向量机 神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]