摘要
[目的/意义]提出一种专利网络快照分析方法,可应用于核心专利预测,进一步发展专利网络分析理论。[方法/过程]首先,利用专利网络快照记录专利技术在不同快照时刻的状态,保留其在发展演化过程中的关键信息;其次,应用图嵌入方法获取记录专利技术演化过程信息的技术关系特征,并开展核心专利预测实验;最后,从不同图嵌入方法、特征表示维度和专利被引时滞等角度,探究技术关系特征对预测效果的影响。[结果/结论]网络快照提供了专利技术演化过程信息,这是静态网络分析方法所不能获取的;利用网络快照和图嵌入方法获取的技术关系特征能够预测核心专利。实证研究发现,技术关系特征的预测效果在一定程度上优于专利指标特征;受专利网络规模的影响,特征表示维度的增大并不会显著提高预测效果,维度在增加至一定数量后预测效果会出现波动;增加专利被引信息可以提高预测的精准性,但会降低预测结果的时效性。
[Purpose/significance]A network snapshot method for core patent prediction is proposed to enrich the existing pa-tent network analysis methods.[Method/process]Firstly,the patented network snapshot is used to record the status of patented technology at different snapshot times,and the key information in the development and evolution process is retained.Secondly,the graph embedding method is applied to the network snapshot to obtain the technical relationship characteristics of recording the infor-mation of patent technology evolution process,and input them into the machine learning classification model to carry out the core patent prediction experiment.Finally,the influence of technical relationship features on prediction results is explored from the per-spectives of different embedding methods,feature representation dimensions and patent citation delay.[Result/conclusion]Net-work snapshots provide information about the evolution process of patented technology,which cannot be obtained by static network analysis methods.The technical relationship features obtained by network snapshot and graph embedding method can predict core pa-tents.Empirical study shows that the prediction effect of technology relationship characteristics is better than that of patent index characteristics to a certain extent.Affected by the scale of patent network,the increase of feature representation dimension will not significantly improve the prediction effect,and the prediction effect will fluctuate after the feature dimension increases to a certain number.The increase of patent citation information can improve the accuracy of prediction.
作者
郭剑明
王婧怡
袁润
Guo Jianming;Wang Jingyi;Yuan Run(Institute of Science and Technology Information of Jiangsu University,Jiangsu Zhenjiang 212013;Library of Jiangsu University,Jiangsu Zhenjiang 212013)
出处
《情报理论与实践》
北大核心
2024年第6期166-174,共9页
Information Studies:Theory & Application
基金
江苏省社会科学基金资助项目“数智驱动下高校图书馆学科服务交互模型及其实现路径研究”的成果之一,项目编号:22TQB001。
关键词
专利网络
网络快照
图表示学习
核心专利
预测模型
patent network
network snapshot
graph representation learning
core patents
prediction model
