科技前沿识别体系中的机器学习应用问题

Application Problems of Machine Learning in Science and Technology Frontier Recognition System

当前科技前沿识别研究方法难以得到更细粒度的分析结果,同时传统计量方法已不能够满足对当前来自网络的开源信息的情报挖掘需求,而机器学习方法可以实现数据细粒度的知识挖掘,因此成为解决科技前沿识别问题的重要手段。对2013—2021年中国知网和Web of Science(WoS)数据库收录的机器学习相关文献,在运用文献计量统计方法进行时间分布、研究主题及热点分析基础上,构建包含数据感知与处理层、情报计算和感知层、情报产品刻画层的开源情报环境下的科技前沿识别体系延伸架构,解读机器学习方法在各层次上的应用问题及关联关系,并提出不同层次需求发展的意见和建议;进而以7 944篇从WoS核心期刊库采集到的“深度学习”主题相关文献作为实验对象,主要针对数据处理中的知识单元构建进行论证。实证结果显示:从应用场景来看,多媒体信息处理的主题热度变化不大,智能机器人的主题热度逐年增高;从机器学习任务来看,目标检测和追踪的主题热度逐年降低,特征工程和数据分类则呈增长趋势。案例分析证明了所提出理论框架的科学性。

Current science and technology frontier identification research method is difficult to get more fine-grained analysis results,at the same time,the traditional measurement method has been unable to meet the current open source information from network mining demand,and machine learning method can realize data fine-grained knowledge mining,therefore become an important means to solve the problem of frontier science and technology identification.On the base of analysis through three aspects of time distribution,research topic and hot spots,on the machine learning related literature included in CNKI and Web of Science(WoS) database from 2013 to 2021,by using the bibliometric statistics method,this paper builds the extension architecture of science and technology frontier identification system under the open source intelligence environment,including data perception and processing layer,intelligence computing and perception layer,and intelligence product characterization layer,interprets the application problems and associations of machine learning methods at all levels,and puts forward opinions and suggestions on the development of different needs at different levels.Furthermore,taking 7 944 literatures related to the topic of "deep learning" from the WoS core journal library as experimental objects,mainly demonstrates the construction of knowledge unit in data processing.From the perspective of application scenarios,the theme heat of multimedia information processing has changed little,and the theme heat of intelligent robot is increasing year by year;from the perspective of machine learning tasks,the topic heat of target detection and tracking decreases year by year,while feature engineering and data classification show an increasing trend.The case analysis proves the scientificity of the proposed theoretical framework.