专利分类序列和文本语义表示视角下的技术融合预测研究  

Technology Convergence Prediction by the Semantic Representation of Patent Classification Sequence and Text

作　　者：张金柱[1] 李溢峰 Zhang Jinzhu;Li Yifeng(Department of Information Management,School of Economics and Management,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学经济管理学院信息管理系,南京210094

出　　处：《情报学报》2022年第6期609-624,共16页Journal of the China Society for Scientific and Technical Information

基　　金：国家自然科学基金面上项目“基于表示学习的专利信息语义融合与深度挖掘研究”(71974095);江苏省研究生科研与实践创新计划项目“基于技术影响力和融合发展性的技术融合关系价值评估研究”(SJCX21_0159)。

摘　　要：为了丰富专利分类的网络和文本语义表示,实现两者更有效的语义融合,提高技术融合预测效果,提出基于专利分类序列和文本语义表示的技术融合预测方法。首先,综合考虑专利分类位置及其上下文语境,直接对专利分类序列进行语义表示,提出基于专利分类序列语义表示的技术融合预测方法;其次,根据专利分类在序列中的重要性排序研究专利分类文本分配方法,形成基于专利分类文本语义表示的技术融合预测方法;在此基础上,设计多种特征融合方法,提出融合专利分类序列结构和文本内容语义表示的技术融合预测方法;最后,基于链路预测的理论和方法对提出的多种技术融合预测方法进行定量评价。在无人机领域的实验证实,专利分类序列语义表示模型的效果明显优于其他网络表示学习方法;依据重要性排序的专利分类文本赋予方式优于文本平均分配方式,基于此的专利分类文本语义表示能更好地进行技术融合预测;“SVM (support vector machine)+哈达玛积”的特征融合方法在所有方法中表现最优,较单一方法均有提高。本文提出的方法能够提高技术融合预测的效果,更好地为技术布局、技术研发提供借鉴和参考。A technology convergence prediction method based on the semantic representation of patent classification sequence and text is proposed to enrich the network and text semantic representation of patent classification, realize their more effective semantic fusion, and improve the effect of technology convergence prediction. First, the semantic representation of the patent classification sequence is directly carried out, and a technology convergence prediction method based on the semantic representation of the patent classification sequence is proposed, considering the location and context of patent classification. Second, the patent classification text allocation method is designed according to the ranking importance of patent classification in the sequence while the technology convergence prediction method is formed based on the semantic representation of patent classification text. Then, a multi-feature fusion method and a technology convergence prediction method combining patent classification sequence structure and the semantic representation of text content are proposed. Finally, based on the theory and method of link prediction, the proposed multi-technology convergence prediction methods are quantitatively evaluated. Experiments in the unmanned aerial vehicle field confirm that the effect of the patent classification sequence semantic representation model is better than other network representation learning methods. The text assignment method of patent classification by importance is better than the average text distribution method, which can better predict technology convergence. In the semantic fusion model,“Support Vector Machine + Hadamard Product”has the best performance, which is better than the single patent classification sequence and the patent classification text method. The method used in this study can better predict the possible technology convergence and provide better reference for technology layout and technology research and development.

关 键 词：技术融合 预测 表示学习 专利分类序列 专利分类文本 

分 类 号：G255.53[文化科学—图书馆学]