情报杂志2025，Vol.44Issue(7)：199-207,9.DOI:10.3969/j.issn.1002-1965.2025.07.024
融合异构图全局结构信息和时间序列的专利价值评估方法
A Patent Value Assessment Method Integrating Global Structural Information of Heterogeneous Graphs and Time Series
摘要
Abstract
[Research purpose]In response to the limitations of existing patent value assessment methods,which fail to fully utilize the global structural information of heterogeneous graphs and the time series of patent citation counts,this paper proposes a method that in-tegrates these features.The aim is to improve patent classification accuracy,particularly in identifying high-value patents.[Research method]This paper presents a patent value assessment model,HNTSM(Heterogeneous Networks and Time Series Model),that in-tegrates global structural information from heterogeneous graphs and time series data.First,a heterogeneous graph neural network model is used to extract background information about patents,incorporating centrality encoding to capture the global structure of the graph.Then,the time series of patent citation counts is integrated,and attention mechanisms are applied to capture the trends in citation count changes.Finally,these features are used for patent value classification.[Research result/conclusion]Experimental results show that the HNTSM model achieves a precision of 77.37％and an F1 score of 76.72％.in identifying high-value patents(A grade)on a U.S.semiconductor patent dataset.Compared to existing methods,the model demonstrates significant improvements,especially through the introduction of global structure information and time series modules,which have a positive impact on patent value assessment outcomes.关键词
专利价值评估/异构图/全局结构信息/时间序列/神经网络模型Key words
patent value assessment/heterogeneous graph/global structural information/time series/neural network model分类
社会科学引用本文复制引用
陈晰,程戈,尹智斌..融合异构图全局结构信息和时间序列的专利价值评估方法[J].情报杂志,2025,44(7):199-207,9.基金项目
湖南省哲学社会科学项目"人工智能赋能知识产权风险预警研究"(编号:22YBA154)研究成果. （编号:22YBA154）
