基于描述能力的视频标题分类

发布时间：2018-02-04 07:00

本文关键词： 视频标题描述能力文本分析视频检索支持向量机　出处：《华南理工大学学报(自然科学版)》2011年07期 　论文类型：期刊论文

【摘要】：在基于文本的视频检索和推荐系统中,视频标题是描述视频内容的必不可少的一个信息来源.然而,人们对视频标题本身的描述能力并没有进行深入的研究.文中根据标题对视频内容的刻画程度将标题的描述能力分为可描述、可理解不可描述和不可理解3个等级,并把标题描述能力的评估问题作为分类问题来处理.鉴于支持向量机(SVM)对小样本分类问题有很好的识别效果,使用SVM作为分类模型.同时,为弥补标题信息的不足,利用视频标题在搜索引擎中的返回结果作为标题的补充信息.采用此方法对汽车领域的5000个视频标题进行分类,结果表明:该方法对可描述类视频标题的识别准确率可达84%;利用标题的搜索结果后,对可描述类和不可理解类标题识别的F值都提高了3%.
[Abstract]:In text-based video retrieval and recommendation systems, video titles are an essential source of information to describe video content. The description ability of video title itself has not been deeply studied in this paper. According to the description degree of video content, the description ability of title is divided into descriptive. There are three levels of comprehensible undescriptive and incomprehensible, and the evaluation of title description ability is treated as a classification problem. In view of the support vector machine (SVM), it has a good recognition effect on the small sample classification problem. Use SVM as the classification model. At the same time, to make up for the lack of title information. The video title returned in the search engine is used as the supplementary information of the title, and the 5,000 video titles in the automobile field are classified by this method. The results show that the recognition accuracy of this method for descriptive video titles can reach 84%. After using the search results of the title, the F value of the description class and the incomprehensible class header is increased by 3%.
【作者单位】：北京理工大学计算机学院;
【基金】：国家自然科学基金资助项目(61003065)
【分类号】：TP391.1
【正文快照】： 视频作为方便高效的记录信息的载体已广泛地应用于人们的日常生活中.手机、数码相机、数码摄录机等视频摄录设备的快速普及使人们倾向于使用视频来捕捉和记录信息,并通过互联网平台进行分享[1].目前,互联网平台中已出现大量的视频共享网站,有些大型网站的视频数量甚至超过千

【参考文献】