立足于企业科技强国战略,推动交叉工程项目发展,深化企业跨领域技术,本文提出一种基于二重BERT(bidirectional encoder representations from transformers)文本分类模型实现企业技术分布匹配的策略.首先,基于深度学习,本文构建了4种BERT模型,结合工程专业标签,对7万条专利文本进行预训练,实现企业属性识别;构建标签张力矩阵,计算加权余弦相似函数,实现技术合作匹配模块,筛选合作者.其次,基于时序分析,实现合作企业间的技术竞合追踪,确定合作程度范围,从"静态"和"动态"角度,为企业跨领域技术合作提出一种定量策略,补充了现有研究针对该问题的系统性、动态性缺陷.最后,选用生物医药工程高成长企业展开实例分析,证实了本文方法的可靠性.
Dynamic and Static Analysis of Cross-Domain Technological Coopetition Based on Dual BERT Text Analysis Approach
Anchored in the strategy of strengthening the nation through corporate science and technology,we promote the development of interdisciplinary engineering projects and the advancement of cross-domain technologies within enterpris-es.This study proposes a strategy for matching the distribution of enterprise technology based on deep learning.Utilizing deep learning,this study constructed four types of BERT(bidirectional encoder representations from transformers)models,combining professional engineering tags to pretrain 70,000 patent texts,thereby identifying corporate attributes.By con-structing a tag tension matrix and calculating the weighted cosine similarity function,a technology cooperation matching module was created to filter collaborators.Additionally,based on a temporal analysis,the technological competition and co-operation between partnering enterprises were tracked,determining the scope of cooperation.Thus,from both'static'and'dynamic'perspectives,a quantitative strategy for cross-domain technological cooperation in enterprises is proposed.The reliability of the method was demonstrated through a case study of a high-growth enterprise in the biopharmaceuticals engi-neering sector.
coopetition analysisenterprise attribute recognitionbinary text categorizationcross-domain cooperationtechnology matching
万方数据
维普
