基于视频弹幕评论的自动摘要生成问题的研究

发布时间：2018-05-27 16:07

本文选题：摘要生成 + 弹幕评论　；参考：《中国科学技术大学》2017年硕士论文

【摘要】：网络共享媒体的蓬勃发展产生了海量的视频数据。同时,大量复杂的视频数据也为视频标记的工作带来了巨大的挑战。自动视频标记技术在这种情况下应运而生,主要用于视频内容的分析并自动获取相关的文字标记。然而,传统方法在为视频数据生成同步文字描述时通常会依赖对应的视频场景和文本信息,导致算法实现起来代价较大。幸运的是,视频弹幕评论提供了关于视频数据的时序信息,也为视频数据的分析提供了新的途径。本文主要研究如何利用弹幕评论信息来为视频数据生成同步的视频描述,这项工作的挑战来自于评论文本中包含较多的噪声数据和复杂的主题信息。针对上述问题,我们基于数据重构原理提出了一种新型摘要生成模型,通过结合弹幕评论之间的时间关联信息,并且考虑在文本层次和主题的层次的联合重构来抽取有代表性的弹幕评论,最终得到的摘要文本将以同步的方式为视频内容提供丰富的描述信息。本文的主要贡献有:(1)提出了基于视频弹幕评论的摘要生成方法来抽取代表性的弹幕评论,并将其应用到视频描述中。(2)通过结合文本和主题的联合重构建立模型来解决弹幕评论中包含的噪音和冗余信息。同时,考虑到问题在时间上的关联信息,相应提出了动态的分段摘要方法。(3)真实数据集上的实验结果证明了我们的方法的有效性,并且验证了利用弹幕评论为视频数据生成文字描述的可行性。
[Abstract]:The flourishing development of network shared media has produced massive video data. At the same time, a large number of complex video data also brings great challenges to the work of video tagging. The technology of automatic video tagging emerges as the times require. It is mainly used for the analysis of video content and the automatic acquisition of relevant text tags. However, the traditional methods usually rely on the corresponding video scene and text information when generating synchronous text description for video data, which leads to the high cost of the algorithm. Fortunately, video screen reviews provide temporal information about video data and a new way to analyze video data. This paper mainly studies how to generate synchronous video description for video data by using project-screen comment information. The challenge of this work is that the comment text contains more noise data and complex subject information. In order to solve the above problems, we propose a new model of summary generation based on the principle of data reconstruction. Considering the joint reconstruction of text level and topic level to extract representative project-screen comments, the resulting summary text will provide rich description information for video content in a synchronous manner. The main contributions of this paper are: (1) A method of abstract generation based on video projectile reviews is proposed to extract representative project-screen reviews. It is applied to the video description. (2) the noise and redundant information contained in the project-screen comments is solved by combining the text and topic reconstruction to build a model. At the same time, considering the time related information of the problem, a dynamic segmenting summary method. 3) the experimental results on the real data set prove the validity of our method. The feasibility of generating text description for video data by projectile commentary is verified.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【相似文献】