图书情报工作 ›› 2021, Vol. 65 ›› Issue (17): 131-141.DOI: 10.13266/j.issn.0252-3116.2021.17.013

• 知识组织 • 上一篇    下一篇

离群专利视角下的新兴技术预测——基于BERT模型和深度神经网络

孔德婧1, 董放2, 陈子婧3, 刘宇涵3, 周源2   

  1. 1. 北京邮电大学现代邮政学院 北京 100876;
    2. 清华大学公共管理学院 北京 100084;
    3. 华中科技大学机械科学与工程学院 武汉 430074
  • 收稿日期:2020-12-21 修回日期:2021-05-31 出版日期:2021-09-05 发布日期:2021-09-01
  • 通讯作者: 周源(ORCID:0000-0002-9198-6586),副教授,博士,博士生导师,通讯作者,E-mail:zhou_yuan@mail.tsinghua.edu.cn
  • 作者简介:孔德婧(ORCID:0000-0002-2575-3514),讲师,博士;董放(ORCID:0000-0003-4271-9702),博士研究生;陈子婧(ORCID:0000-0001-7761-5810),硕士研究生;刘宇涵(ORCID:0000-0002-3574-8479),硕士研究生。
  • 基金资助:
    本文系国家自然科学基金项目"基于多源知识图谱的产业融合路径及机制研究"(项目编号:72004016)和国家自然科学基金项目"基于多源异构网络视角的新兴产业创新扩散作用机制及政策研究"(项目编号:71974107)研究成果之一。

Prediction of Emerging Technologies from the Perspective of Outlier Patents——Based on Bert Model and Deep Neural Networks

Kong Dejing1, Dong Fang2, Chen Zijing3, Liu Yuhan3, Zhou Yuan2   

  1. 1. School of Modern Post, Beijing University of Posts and Telecommunications, Beijing 100876;
    2. School of Public Policy and Management, Tsinghua University, Beijing 100084;
    3. School of Mechanical Science and Engneering, Huazhong University of Science and Technology, Wuhan 430074
  • Received:2020-12-21 Revised:2021-05-31 Online:2021-09-05 Published:2021-09-01

摘要: [目的/意义] 由于新兴技术本身的超前性,其刚出现的关注度往往不是很高。目前研究更多遵循技术发展路径依赖进行新兴技术的识别,会忽略一些颠覆现有技术轨道的技术研发。通过对与领域内主流技术相似度较低的离群专利进行分析,可以更有效地识别这类技术研发并预测新兴技术。[方法/过程] 提出一种基于深度学习的离群专利识别与新兴技术预测方法。首先使用BERT预训练模型基于专利文本构建相似度网络,识别离群专利,然后基于DNN模型构建离群专利指标与技术影响力之间的关系,实现从海量离群专利中快速、准确地预测新兴技术。最后以数控系统领域为例,从德温特专利数据库获取近10年领域内所有专利,进行实证分析。[结果/结论] 数控系统领域的实证分析结果验证了模型的有效性,同时对国家的技术发展政策制定以及相关领域企业技术布局具有重要的指导意义。

关键词: 新兴技术, 深度学习, 离群专利, 数控系统

Abstract: [Purpose/significance] Due to the advanced nature of emerging technologies, they are often marginalized at the initial stage of formation. Most of present researches forecast emerging technologies by analyzing the mainstream technology development path, which would neglect some research that disrupts existing technology routes. By analyzing outlier patents that are less similar to the mainstream technologies in the field, it can identify and forecast the future emerging technologies more effectively.[Method/process] This paper presented an outlier patent identification and emerging technology prediction method based on deep learning. Firstly, the Bert pre-trained model was used to construct the similarity network based on texts of patents and outlier patents identification. The relationship model between outlier patent indicators and technical influence was then built based on DNN model, thus realizing the fast and accurate emerging technology prediction using large-scale outlier patents. Finally, an empirical analysis was conducted in the field of numerical control system with all patents applied in the last ten years obtained from DI database.[Result/conclusion] The result of empirical analysis in the field of numerical control system not only verifies the validity of the model, but also has important guiding significance to the formulation of national technology development policy and the technology layout of enterprises in related fields.

Key words: emerging technologies, deep learning, outlier patents, numerical control system

中图分类号: