项目名称: 基于概率图模型的海量可视媒体协同理解与推荐研究
项目编号: No.61305018
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 肖宪
作者单位: 中国科学院自动化研究所
项目金额: 25万元
中文摘要: 随着互联网可视媒体数据呈爆炸性增长,海量可视媒体数据的理解与推荐成为国际难题。信息"多"又"杂"是互联网可视媒体资源的特色,用户很难找到所需的可视媒体信息,可视媒体理解与推荐是解决该问题的重要途径,因此,我们选择基于概率图模型的海量可视媒体理解与推荐理论和方法研究。首先,针对海量异构的互联网可视媒体信息,借助概率图模型理论和超网络结构方法,研究可视媒体、用户关系网和相关文字语义概念在融合特征空间中的网络化表示,提出可视媒体对象协同语义关联网的自动构建和快速推理方法;其次,依据协同语义关联网和概率图模型理论,研究对存在关联关系的互联网可视媒体数据理解算法,以及推演到不存在关联关系的可视媒体数据的算法,丰富协同语义关联网;最后,从协同语义关联网中,研究抽离针对用户的个性化子关联网方法,实现对用户的可视媒体个性化推荐,并在国际通用的地标建筑和服装多媒体数据库中验证项目研究成果的正确性、有效性。
中文关键词: 概率图模型;可视媒体理解;可视媒体推荐;可视媒体语义关联网;语义相关性
英文摘要: The billions of visual media in large scale photo collections offer both exciting opportunities and significant challenges for computer vision and for the area of visual media understanding and recommendation in particular. The huge amount and the complex relationship make the large scale visual media understanding and recommendation becomes a worldwide challenging task. It is hard to manage the large scale visual media effectively. Meanwhile, it is also hard for the internet users to obtain helpful visual media. Visual media understanding and recommendation is an important way to handle the above problems. Therefore, in this proposal, we consider the visual media understanding and recommendation problems based on probabilistic graphical model (PGM). Firstly, the visual features, user relationship and semantic concept are represented in the fusion feature space. The automatic construction and fast reasoning method for the fusion feature space is proposed based the probabilistic graphical model. Secondly, the proposed method is utilized to analyze the visual media with the incidence relationship and reasoning the visual media without the incidence relationship. After that, an abundant collaborative semantic association network for visual media is obtained. Finally, the collaborative semantic association network o
英文关键词: Probabilistic Graphical Model;Visual Media Understanding;Visual Media Recommendation;Visual Media Semantic Association Network;Semantic Correlation