摘要
[目的/意义]在技术相似度的基础上,引入了多层异构网络链路预测算法,提出一种产学研企业与学者专利合作关系预测的方法。[方法/过程]首先,构建了企业与学者的特征向量体系,构建节点特征;其次,构建学者合作网络、企业-学者合作网络两层网络,采用R-GCN算法进行模型训练,同时计算企业学者之间技术相似度;最后,将技术相似度与R-GCN模型结果进行逻辑回归拟合,得到最终合作预测结果。[结果/结论]通过对生物医药领域专利数据集进行实证分析,技术相似度+多层异构网络链路预测算法测试集的准确率、召回率、AUC、MSE、F2-Score、MRR分别为67.74%、78.02%、90.95%、4.41%、34.00%、52.05%,其中MRR相比R-GCN、Jaccard、Cosine、Euclidean分别提升了8.67倍、1.78倍、2.58倍、1.65倍。说明本文构建的融合多层异构网络链路预测与技术相似度的算法是有效且具可行性的,在产学研合作关系的预测和合作学者的推荐中具有理论意义与实践价值。
[Purpose/Significance]On the basis of technical similarity,a multi-layer heterogeneous network link prediction algorithm is introduced,and a method for predicting the patent cooperation relationship between industry university research enterprises and scholars is proposed.[Method/Process]Firstly,the feature vector system of enterprises and scholars was constructed,and the node features were constructed.Secondly,a two-layer network of scholar cooperation network and enterprise scholar cooperation network was built,R-GCN algorithm for model training was used,and the technical similarity between enterprise scholars was calculated.Finally,the technical similarity was fitted with the results of R-GCN model by logistic regression,and the final cooperative prediction results were obtained.[Result/Conclusion]Through the empirical analysis of the patent data set in the field of biomedicine,the accuracy,recall,AUC,MSE,F2 score and MRR of the technical similarity+multi-layer heterogeneous network link prediction algorithm were 67.74%、78.02%、90.95%、4.41%、34.00%、52.05%,respectively,among which the MRR was 8.67 times、1.78 times、2.58 times、1.65 times higher than that of R-GCN,Jaccard,cosine and Euclidean respectively.[Limitations]Adding new nodes in R-GCN model requires global training and calculation.It is suitable for small networks and static networks.[Conclusion]This scheme proposes a scheme to directly recommend patent cooperation scholars for enterprises,which integrates patent content and patent cooperation network,and has a good prediction effect.
作者
王骞敏
鄢春根
闵超
Wang Qianmin;Yan Chungen;Min Chao(Library,Hangzhou Normal University,Hangzhou 311100,China;Ceramic Intellectual Property Information Center,Jingdezhen Ceramic University,Jingdezhen 333403,China;School of Information Management,Nanjing University,Nanjing 210023,China)
出处
《现代情报》
CSSCI
2023年第5期54-65,共12页
Journal of Modern Information
基金
江苏省社会科学基金青年项目“基于专利数据挖掘的企业技术创新活动研究”(项目编号:18TQC005)
浙江省教育厅一般项目“产学研深度融合背景下高校专利导航服务模式构建”(项目编号:Y202248407)。
关键词
产学研
专利合作
R-GCN
多层异构网络
技术相似度
Industry-University-Research
patent cooperation
R-GCN
multi-layer heterogeneous network
technical similarity
