项目名称: 社交网络级联数据流异常检测模型研究
项目编号: No.61502479
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 周川
作者单位: 中国科学院信息工程研究所
项目金额: 20万元
中文摘要: 社交网络级联数据(Cascade data)是社交网络中信息传播所留下的传播轨迹数据,级联数据异常检测旨在检测出异常的信息传播轨迹(比如爆发性的传播轨迹)。与传统基于向量数据的检测方法不同,级联数据属于图数据(Graph data)类型,因此传统的向量数据检测模型都不能直接使用在级联数据上。为此,本项目将针对社交网络级联数据异常检测这个新问题,从级联数据的特征抽取,级联数据的异常检测模型,以及大规模传播级联数据的快速求解算法这三个问题出发,具体开展以下研究内容:1)级联数据基于传播子图结构(Subgraphs)的特征抽取方法;2)级联数据基于时序概率图模型的聚类异常检测模型;3)大规模级联数据基于粗化的快速求解算法。 本项目的研究成果对社交网络商业广告异常传播模式发现、 社交网络舆情分析等领域具有重要的理论价值和应用前景。
中文关键词: 级联数据;异常检测;图特征提取与分类;概率图模型;图粗化算法
英文摘要: Social network cascade data is the propagation trajectory data to describe the information propagation in social networks. Cascade data anomaly detection aims to detect abnormal information propagation path (e.g. bursty propagation path). Different from the traditional detection method based on vector data, cascade data is graph data, and the traditional detection model can not be used directly. To this end, this project will address the new problem of cascade data anomaly detection. From the issues of feature extraction, anomaly detection model and fast algorithm, we will carry out the following researches: 1) feature extraction method based on the cascaded data content and the structure of diffusion; 2) anomaly detection model with classification and clustering; 3) cascade data coarsen and fast algorithm for large-scale network. The research results of this project have important theoretical value and application prospect in the fields of social network analysis, such as public opinion on social networks and commercial advertising pattern discovery.
英文关键词: cascade data;anomaly detection;graph features extraction and classification;probabilistic graphical model;graph coarsening algorithm