ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2014, Vol. 51 ›› Issue (12): 2733-2745.doi: 10.7544/issn1000-1239.2014.20131337

• 信息处理 • 上一篇    下一篇



  1. (服务计算技术与系统教育部重点实验室(华中科技大学) 武汉 430074) (集群与网格计算湖北省重点实验室(华中科技大学) 武汉 430074) (
  • 出版日期: 2014-12-01
  • 基金资助: 

Online Video Advertising Based on Fine-Grained Video Tags

Lu Feng, Wang Zirui, Liao Xiaofei,Jin Hai   

  1. (Key Laboratory of Services Computing Technology and System (Huazhong University of Science and Technology), Ministry of Education, Wuhan 430074) (Key Laboratory of Cluster and Grid Computing of Hubei Province (Huazhong University of Science and Technology), Wuhan 430074)
  • Online: 2014-12-01

摘要: 随着互联网的发展,对精彩视频点进行标注、评论和分享成为趋势.这类群体智慧信息的有效利用将有助于提升视频广告的投放效果.首先将用户提供的细粒度视频标签收集起来,通过视频时间轴加权计算生成视频热点,进而利用视频热点描述信息基于分类匹配的思想来选取广告,最后找出视频热点内用户对视频关注度下降幅度最大的时间点投放广告.实验证明,在数量为百万级的视频集合中,该方法选取的广告与视频的相关性达到85%左右.用户在广告播放过程中关闭广告的概率小于10%.与目前广泛应用的广告投放方式相比,广告的平均播放时间能提升21.5%,广告点击率能从0.65%提高至0.73%.

关键词: 在线视频广告, 细粒度视频标签, 视频热点, 机器学习, 定向广告

Abstract: With the development of the Internet, it has been a trend of manual tagging, labeling and sharing videos. Rational use of these swarm intelligence will help to improve the effectiveness of video advertising. The method presented in this paper first collects the fine-grained user video tags, and generates the video hotspots by the video timeline-weighted method. Then, based on the idea of the classification matching, the description of the video hotspots can be used to select the advertising. At last, the time points that the popular attention has dropped by the biggest level are found to put advertising. Experiments show that, among the mega-scale video set, the content correlation between the hotspot and the advertisements selected by this method can reach 85%. The probability that the users close ads windows is less than 10%. Compared with the ads system that has been widely adopted so far, the average broadcast time of the new method can be increased by 21.5%, the click-through rate is improved from 0.65% to 0.73%.

Key words: online video advertising, fine-grained video tags, video hotspots, machine learning, target advertising