[Purpose/significance] A good key technology identification method can provide better support for key technology identification, prediction and research and development at all levels. [Method/process] In this paper, a key technology identification method based on Bert-LDA was proposed, which combined BERT and LDA to make up for the lack of contextual semantic information in a single LDA topic model. An empirical study was carried out with agricultural robots as an example. Specifically, it included the following processes: ① Constructing BERT semantic feature vector and LDA topic feature vector based on Python, combining them in a high-dimensional space, and learning the low-dimensional latent space representation of the concatenated vector by using an autoencoder; ② In the potential space representation, K-means algorithm was used to realize semantic association clustering, and the effect diagram of two-dimensional clustering and key technology subject word cloud maps were drawn; ③ Determining key technologies; ④ In the field of agricultural robots, the effectiveness of this method was demonstrated by comparing with the results of TI patent analysis and the list of key generic technologies for agricultural equipments in the "Made in China 2025" technology roadmap for key areas. [Result/conclusion] The results show that the Bert-LDA model improves the coherence of topic clustering and the accuracy of fine-grained classification. With a good key technology identification accuracy and recall rate, there are good inclusiveness, compatibility and adaptability to the identified literature data sets of different databases and publishing types. It can be widely used to identify all kinds of key technologies.
[1] ALTUNTAS S, DERELI T, KUSIAK A. Forecasting technology success based on patent data[J]. Technological forecasting and social change, 2015, 96(7):202-214.
[2] PARK I, PARK G, YOON B, et al. Exploring promising technology in ICT sector using patent network and promising index based on patent information[J]. ETRI journal, 2016, 38(2):405-415.
[3] LEE C, KWON O, KIM M, et al. Early identification of emerging technologies:a machine learning approach using multiple patent indicators[J]. Technological forecasting and social change, 2018, 127(2):291-303.
[4] LIU X, PORTER A L. A 3-dimensional analysis for evaluating technology emergence indicators[J]. Scientometrics, 2020, 124(1):27-55.
[5] 江娴, 魏凤. 基于专利分析的共性技术识别研究框架[J]. 情报杂志, 2015, 34(12):79-84.
[6] 杨武, 杨大飞. 基于专利数据的产业核心技术识别研究——以5G移动通信产业为例[J]. 情报杂志, 2019, 38(3):39-45,52.
[7] 宋欣娜, 郭颖, 席笑文. 基于专利文献的多指标新兴技术识别研究[J]. 情报杂志, 2020, 39(6):76-81,88.
[8] CHO T S, SHIH H Y. Patent citation network analysis of core and emerging technologies in Taiwan:1997-2008[J]. Scientometrics, 2011, 89(3):795-811.
[9] HO M H C, LIN V H, LIU J S. Exploring knowledge diffusion among nations:A study of core technologies in fuel cells[J]. Scientometrics, 2014, 100(1):149-171.
[10] KUUSI O, MEYER M. Anticipating technological breakthroughs:using bibliographic coupling to explore the nanotubes paradigm[J]. Scientometrics, 2007, 70(3):759-777.
[11] YOU H, LI M, HIPEL K W, et al. Development trend forecasting for coherent light generator technology based on patent citation network analysis[J]. Scientometrics, 2017, 111(1):297-315.
[12] 李蓓, 陈向东. 基于专利引用耦合聚类的纳米领域新兴技术识别[J]. 情报杂志, 2015, 34(5):35-40.
[13] 杨艳萍, 董瑜, 韩涛. 基于专利共被引聚类和组合分析的产业关键技术识别方法研究——以作物育种技术为例[J]. 图书情报工作, 2016, 60(19):143-148,124.
[14] CHEN H, ZHANG G, ZHU D, et al. Topic-based technological forecasting based on patent data:a case study of Australian patents from 2000 to 2014[J]. Technological forecasting and social change, 2017, 119(6):39-52.
[15] YANG C, ZHU D, WANG X, et al. Requirement-oriented core technological components' identification based on SAO analysis[J]. Scientometrics, 2017, 112(3):1229-1248.
[16] ZHOU Y, DONG F, LIU Y, et al. Forecasting emerging technologies using data augmentation and deep learning[J]. Scientometrics, 2020, 123(1):1-29.
[17] 李欣, 王静静, 杨梓, 等. 基于SAO结构语义分析的新兴技术识别研究[J]. 情报杂志, 2016, 35(3):80-84.
[18] 周源, 刘宇飞, 薛澜. 一种基于机器学习的新兴技术识别方法:以机器人技术为例[J]. 情报学报, 2018, 37(9):939-955.
[19] 陈伟, 林超然, 孔令凯,等. 基于专利文献挖掘的关键共性技术识别研究[J]. 情报理论与实践, 2020, 43(2):92-99.
[20] DEVLIN J, CHANG M W, LEE K, et al. Bert:pre-training of deep bidirectional transformers for language understanding[J]. arXiv preprint, 2018, arXiv:1810.04805.
[21] ASGARI-CHENAGHLU M, FEIZI-DERAKHSHI M R, FARZINVASH L, et al. TopicBERT:a cognitive approach for topic detection from multimodal post stream using BERT and memory-graph[J]. Chaos, solitons & fractals, 2021, 151(10):111274.
[22] THOMPSON L, MIMNO D. Topic modeling with contextualized word representation clusters[J]. arXiv preprint, 2020, arXiv:2010.12626.
[23] ABUZAYED A, AL-KHALIFA H. BERT for Arabic topic modeling:An experimental study on BERTopic technique[J]. Procedia computer science, 2021, 189(11):191-194.
[24] 付静,龚永罡,廉小亲,等.基于BERT-LDA的新闻短文本分类方法[J].信息技术与信息化,2021(2):127-129.
[25] 庄穆妮,李勇,谭旭,等.基于BERT-LDA模型的新冠肺炎疫情网络舆情演化仿真[J].系统仿真学报,2021,33(1):24-36.
[26] 李越,毛存礼,余正涛,等.融合主题及上下文特征的汉缅双语词汇抽取方法[J].小型微型计算机系统,2021,42(1):91-95.
[27] BLEI D M, NG A Y, JORDAN M I. Latent dirichlet allocation[J]. The journal of machine learning research, 2003, 3(1):993-1022.
[28] MCINNES L, HEALY J, MELVILLE J. Umap:uniform manifold approximation and projection for dimension reduction[J]. arXiv preprint, 2018, arXiv:1802.03426.
