多模态健身运动知识图谱构建  

Research on Constructing a Multi-Modal Knowledge Graph for Scientific Exercise

作　　者：余卓成 刘心怡 李素建 赵冰婵 罗文 杨彤晨 丁立祥[2] 宋继鹏 YU Zhuocheng;LIU Xinyi;LI Sujian;ZHAO Bingchan;LUO Wen;YANG Tongchen;DING Lixiang;SONG Jipeng(School of Computer Science,Peking University,Beijing 100871,China;Beijing Shijitan Hospital,Beijing 100038,China)

机构地区：[1]北京大学计算机学院,北京100871 [2]北京世纪坛医院,北京100038

出　　处：《首都体育学院学报》2024年第5期481-489,共9页Journal of Capital University of Physical Education and Sports

基　　金：国家重点研发计划项目(2022YFC3600400)。

摘　　要：构建健身运动知识图谱可以为人们科学运动提供直观便捷的指导,对推动健身研究领域的发展和促进人们身体健康具有重要意义。而如何全面获取科学健身运动的多模态知识,并进行系统的关联和集成以及形成多模态知识图谱,目前相关研究较少,且面临着诸多挑战。针对上述问题,利用自然语言处理、深度学习技术、视觉识别技术等,以人机结合的方式探索构建多模态科学健身运动知识图谱。具体步骤包括:1)数据选择和处理,确保知识图谱的基础数据可靠且全面;2)设计多模态知识体系,以整合文字、图像等多种信息形式;3)运用关键技术获取知识,特别是将运动视频转换为细粒度运动文本进行描述;4)构建可视化平台,使知识图谱更加直观和易于使用。通过系统地关联和集成多模态的健身知识,构建了一个包含丰富信息的健身运动知识图谱。该图谱系统不仅能够有效地将运动视频转换为详细的文本描述,还能可视化地直观展示图片、视频等信息,为用户提供易于理解和使用的科学运动指导。多模态科学健身运动知识图谱的构建,为其在健身方面的应用提供了重要参考。这表明,多模态知识的获取和集成是实现科学健身指导的重要途径,能够有效推动健身研究领域的发展以及促进人们的身体健康。Constructing an exercise knowledge graph can provide intuitive and convenient guidance for scienti-fic exercise,significantly contributing to the development of the fitness course and the promotion of public health.However,there are currently few studies on how to comprehensively acquire multi-modal knowledge of scientific exercises and systematically integrate this knowledge to form a multi-modal knowledge graph.This task faces numerous challenges.To address these issues,we explore the construction of a multi-modal scientific fitness exercise knowledge graph using natural language processing,deep learning,and visual recognition techn-ologies in a human-machine collaborative manner.The specific steps include:1)data selection and processing to ensure the reliability and comprehensiveness of the foundational data for the knowledge graph;2)designing a multimodal knowledge system to integrate various forms of information,such as text,images,and videos;3)empl-oying key technologies to acquire knowledge,particularly converting exercise videos into fine-grained textual descriptions;4)building a visualization platform to make the knowledge graph more intuitive and user-friendly.By systematically relating and integrating multi-modal fitness knowledge,we have successfully constructed a fitness exercise knowledge graph rich in information.This graph not only effectively converts exercise videos into detailed textual descriptions but also intuitively displays the information through a visualization platform,providing users with easy-to-understand and user-friendly scientific exercise guidance.The construction of the multi-modal scientific exercise knowledge graph provides an important reference for the construction and appli-cation of knowledge graphs in the fitness field.This indicates that acquiring and integrating multi-modal knowledge is an essential approach to achieving scientific fitness guidance,which can effectively promote the development of the fitness course and enhance people's health.

关 键 词：多模态知识图谱 知识获取 科学运动 

分 类 号：G80-05[文化科学—运动人体科学]