摘要
针对采集视频数据时存在数据冗余、资源占用多、采集质量低、遗漏关键信息等问题,研究人员提出一种视频多模态数据采集模型.首先,从资源服务器获取热度影响因子及视频数据,将视频数据解析为图像、文本、音频等多模态数据,并建立多模态视频数据与热度影响因子的时间关联矩阵.其次,通过“熵权法”计算“关注数”、“点赞数”、“弹幕数”、“分享数”及“在看数”5个热度影响因子的权重值,实现视频热度的量化.而后,提出了多分段旋转门(MSDT)数据采集策略,基于视频热度值自适应调整数据采集时间间隔,在视频热度上升、下降及高热度持续阶段增加采集量;反之,降低采集量,在保证采集质量的同时兼顾采集效率.最后,通过对比实验验证所提策略的可行性,结果表明,MSDT能够实现分段动态调整采集频率,降低数据采集误差约24%,降低数据采集量约26%,提高采集效率17%左右.
Aiming at the problems of data redundancy,high resource occupation,low acquisition quality,and missing key information when collecting video data,researchers proposes a video multimodal data acquisition model.First,the heat influence factors and video data are obtained from the resource servers,analyzed into multimodal data such as image,text and audio,and the time correlation matrix between multi-modal data and heat influence factor is established.Second,the entropy weight method is used to calculate the weight value of five popularity influence factors which are the number of"concerns","likings","barrages","sharings"and"viewings"to realize the quantification of video popularity.Then,A Multi-segment Swing Door Trending(MSDT)data acquisition strategy is proposed,which adaptively adjusts the data acquisition time interval based on the video heat value and increases the amount of data acquisition in the stages of video heat rise,fall and high heat duration and reduces the amount of collection conversely,which ensures the collection quality while considering the collection efficiency.Finally,the feasibility of the proposed strategy is verified by comparative experiments.The results show that MSDT can adjust the acquisition frequency in segments dynamically,reduce the data acquisition error by about 24%and the amount of data acquisition by about 26%,and improve the overall acquisition efficiency by about 17%.
作者
肖萍
刘荆欣
王妍
臧洁
XIAO Ping;LIU Jing-xin;WANG Yan;ZANG Jie(College of Public Security Information Technology and Intelligence,Criminal Investigation Police University of China,Shenyang 110854,China;College of Information,Liaoning University,Shenyang 110036,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2023年第2期383-391,共9页
Journal of Chinese Computer Systems
基金
国家重点研发计划项目(2019YFB1405804)资助.
关键词
多模态
视频数据采集
热度因子
自适应策略
多分段旋转门算法
multimodal
video data collection
heat factor
adaptive strategy
multi-segment swing door trending
