基于融合新闻影响力和图注意力网络聚类方法的颠覆性技术识别  

Disruptive Technology Identification Based on News Influence and Enhanced Graph Attention Network Clustering Method

作　　者：王良 李牧南[1] Wang Liang;Li Munan(School of Business Administration,South China University of Technology,Guangzhou 501641)

机构地区：[1]华南理工大学工商管理学院,广州510640

出　　处：《图书情报工作》2024年第15期27-43,共17页Library and Information Service

基　　金：国家社会科学基金重点项目“加快我国科技自立自强发展战略问题研究”(项目编号:22AZD035)研究成果之一。

摘　　要：[目的/意义]许多现有研究依赖于专利数据来识别颠覆性技术,但这些方法在专利文本的主题聚类分析方面仍有进一步优化的潜力。[方法/过程]通过构建基于新闻影响力增强的图注意力网络,以及自适应分配注意力权重,有效捕捉和充分利用技术主题词共现网络节点关系,在生成有代表性的节点向量后再进行专利文本主题聚类分析,可以进一步辅助识别潜在的颠覆性技术。[结果/结论]为进一步验证方法的有效性,选择智慧城市和工业互联网两个新兴技术领域进行实证检验。理论和实证分析显示,这种融合新闻影响力的图注意力网络聚类方法可以进一步丰富当前有关颠覆性技术识别的方法体系。[Purpose/Significance]Many existing studies rely on patent data to identify disruptive technologies,but these methods still have potential for further optimization in terms of topic clustering analysis of patent texts.[Method/Process]By constructing a graph attention network based on the enhancement of news influence and adaptive allocation of attention weights,the node relationship of the co-occurrence network of technical subject words was effectively captured and fully utilized.After generating representative node vectors,it conducted thematic clustering analysis of patent text,which could further assist in identifying potential disruptive technologies.[Result/Conclusion]In order to further verify the effectiveness of the method,it selects two emerging technology fields of smart city and industrial Internet for empirical testing.The theoretical and empirical analysis shows that this clustering method of graph attention network,which integrates the influence of news,can further enrich the current methodological system on the identification of disruptive technologies.

关 键 词：颠覆性技术 技术识别 主题聚类 新闻数据 图注意力网络 

分 类 号：G250[文化科学—图书馆学] TP393[自动化与计算机技术—计算机应用技术]