摘要
[目的/意义]技术机会预测有利于国家和企业管理者识别技术未来的发展方向,从而调整技术发展战略,在技术竞争中占据优势地位。[方法/过程]提出一种基于深度学习的技术机会预测方法。首先运用AP(affinity propagation)聚类算法实现对技术领域的主题划分。其次运用Doc2Vec算法计算出各技术领域专利文本相似度情况,进而识别出具有发展潜力的技术领域。再次采用生成式拓扑映射(generative topographic mapping,GTM)算法对发展潜力技术领域绘制专利地图,通过GTM逆向映射获得技术机会。最后,构建基于深度学习的链接预测模型,对识别出的技术机会进行链接预测,从而获得高发展概率的技术机会。[结果/结论]使用新能源汽车专利数据对方法的有效性进行验证,结果显示基于深度学习的链接预测模型的预测准确率、召回率和F1值均优于其他预测模型,并对新能源汽车的技术机会进行预测。
[Purpose/significance]Technology opportunity prediction helps national and enterprise managers to identify the future direction of technology development,to adjust the development strategy and occupy a favorable posture for technology competition.[Method/process]In this paper,a deep learning-based technology opportunity prediction method was proposed.Firstly,the Affinity Propagation(AP)clustering algorithm was applied to achieve the subject classification of technology fields.Secondly,the Doc2Vec algorithm was used to calculate the similarity of patent texts in each technology area,and then identify the technology areas with development potential.Thirdly,the Generative Topographic Mapping(GTM)algorithm was used to map the patent areas with development potential,and we got the technology opportunities through GTM inverse mapping.Finally,a link prediction model based on deep learning was constructed to predict the links of the identified technology opportunities,to obtain the technology opportunities with high development probability.[Result/conclusion]This paper uses new energy vehicle patent data to verify the effectiveness of the method.The results show that the prediction accuracy,recall and F1 value of the deep learning-based link prediction model outperform other prediction models and predict the technology opportunities for new energy vehicles.
作者
桂美增
许学国
Gui Meizeng;Xu Xueguo(Accounting college,Zhejiang University of Finance and Economics,Hangzhou 310018;School of management,Shanghai University,Shanghai 200444)
出处
《图书情报工作》
CSSCI
北大核心
2021年第19期130-141,共12页
Library and Information Service
基金
国家自然科学基金青年项目"关键利益相关者视角下新兴产业创新政策作用机制与仿真优化:以新能源汽车为例"(项目编号:71704101)研究成果之一。
关键词
技术机会预测
深度学习
链接预测
生成式拓扑映射
technology opportunity prediction
deep learning
link prediction
the generative topographic mapping
