摘要
社会媒体网络中不仅包含了用户、文本、图片和视频等多种模态的数据,还包含了反映不同模态数据之间交互的群体特征。为了更好地描述社会媒体网络,从而为上层应用提供更好的服务,提出了一种基于深度神经网络的社会媒体网络模型。该模型采用深度神经网络对单个模态的数据进行学习,从而得到任意一个模态数据的潜在特征表示方法。对于两种不同模态的数据,利用具有高斯分布的先验矩阵与两个模态数据的后验分布建立反映这两个模态数据间群体特征的生成模型。实验结果表明,提出的模型在网络结构的链接分析中具有更好的预测效果,能有效地描述社会媒体网络的整体特征。
Social media networks include not only multi-modal data, such as users, text, images, video, and so on, but also collective effect between data of different modals. In order to better describe social media networks and provide bet- ter services for applications, this paper proposed a deep neural network based social media network model. The proposed model models data of each single modal with a deep neural network, and then gets a latent feature representation for each modal data. For data between two different modals, a Gaussian distributed prior matrix and two posterior distribu- tions of different modals were applied to build a generative model that describes the collective effect between two diffe- rent modals. The experiments show that the proposed model has better prediction performance in link analysis of net- works than related works,and can describe the whole underlying social media network effectively.
出处
《计算机科学》
CSCD
北大核心
2016年第4期252-255,263,共5页
Computer Science
基金
国家自然科学基金资助项目(61303074)资助
关键词
深度学习
神经网络
社会媒体网络
特征学习
链接预测
Deep learning, Neural networks, Social media network, Feature learning, Link prediction