基于SimCSE-LDA和异常检测的颠覆性技术识别方法——以农业机器人为例  

Disruptive Technology Identification Method Based on SimCSE-LDA and Anomaly Detection:A Case Study of Agricultural Robots

作　　者：王秀红[1,2] 王欣 王少凡 刘浩东 张宗毅 Wang Xiuhong

机构地区：[1]江苏大学科技信息研究所,江苏镇江212013 [2]江苏大学图书馆,江苏镇江212013 [3]江苏大学中国农业装备产业发展研究院,江苏镇江212013

出　　处：《情报理论与实践》2023年第5期135-143,共9页Information Studies:Theory & Application

基　　金：国家重点研发计划项目“农业装备制造产业集聚区域网络协同制造集成技术研究与应用示范”的成果之一,项目编号:SQ2020YFB170242。

摘　　要：[目的/意义]识别颠覆性技术有助于在制定研发和市场战略布局中获得首发优势,如何在大数据背景下综合提高颠覆性技术识别的精准率、召回率和识别效率至关重要。[方法/过程]结合机器学习与专家知识,首先,构建SimCSE-LDA语义相似度表征模型,以深入挖掘专利摘要中隐含的关键技术主题,实现更深层次的显著特征主题识别,并通过聚类效果评价指标检验其识别效果,进一步结合专家知识判定关键技术主题名称,对主题间内在联系判定,最终识别出关键技术。其次,运用突变性表征颠覆性技术内在特质,基于CBLOF算法对关键技术主题进行异常检测,将计算得到的技术主题异常分数作为判断技术突变程度的依据,从而识别出颠覆性技术。最后,结合领域专家知识和《中国制造2025》验证颠覆性技术识别效果。[结果/结论]以农业机器人为例,以德温特专利数据库的DWPI英文改写专利摘要文本为数据源,进行实证分析,验证了该颠覆性技术识别方法的可行性与有效性。[Purpose/significance]Identifying disruptive technologies can help to gain a first-mover advantage in developing R&D and market strategy layout,so it is crucial to improve the precision,recall and efficiency of disruptive technology identification in the context of big data in a comprehensive manner.[Method/process]This paper combined machine learning and expert assessment:firstly,SimCSE-LDA semantic similarity representation model was built to dig deep into the key technology topics implied in patent abstracts to achieve deeper identification of salient feature topics,and demonstrated its identification effectiveness by clustering effect evaluation indexes.Then combined with expert knowledge to determine the key technology topic names,and to identify the key technologies according to the intrinsic connection between the topics.Secondly,mutability was used to characterize the inherent characteristic of disruptive technologies,and the CBLOF algorithm was used to detect the anomaly of key technology topics.Then the calculated anomaly scores of technology topics were used as the basis to judge the degree of technological mutation,so as to identify disruptive technologies.Finally,the effectiveness of disruptive technology identification was verified by combining domain expert knowledge and“Made in China 2025”.[Result/conclusion]An empirical analysis was conducted in the field of agricultural robots,based on DWPI English rewritten patent abstracts from Derwent Innovation Index as the data source,and the result verifies the feasibility and effectiveness of the disruptive technology identification method.

关 键 词：颠覆性技术 SimCSE-LDA模型 异常检测 技术识别 农业机器人 

分 类 号：S24[农业科学—农业电气化与自动化] TP242[农业科学—农业工程]