富属性异质信息网络的可约束异常检测
发布时间：2019-04-23 19:15
【摘要】：针对仅考虑网络结构来对异质信息网络进行异常点发现可能带来的结果失真、难以理解等问题,提出一种富属性异质信息网络的可约束异常检测算法.通过将信息丰富的交互数据建模成富属性异质信息网络,以带属性元路径来指定用户感兴趣的属性和子空间,综合网络结构和属性内容两方面来评估节点的异常度,给出了可约束的异常检测算法框架.在Arxiv真实数据集上进行了实验,以带属性元路径来指定对作者、论文及论文的标题和摘要等方面的约束,对多个查询输出了异常度从高到低的节点列表及约束域异常点集合.结果表明:相比仅考虑网络结构或仅考虑属性内容的基准算法,平均准确率提高12.95%以上.
[Abstract]:In this paper, a constrained anomaly detection algorithm for heterogeneous information networks with rich attributes is proposed, aiming at the distortion and difficulty of finding outliers in heterogeneous information networks considering only the network structure. By modeling the information-rich interactive data into an attribute-rich heterogeneous information network, using attribute element paths to specify the attributes and subspaces of interest to the user, the anomaly degree of the nodes is evaluated by integrating the network structure and attribute content. The framework of constrained anomaly detection algorithm is presented. Experiments are carried out on Arxiv real data sets. Constraints on authors, papers, titles and abstracts are specified with attribute element paths, and outliers from high to low in node list and constraint domain outliers are outputted for multiple queries. The results show that the average accuracy of the benchmark algorithm is 12.95% higher than that of the benchmark algorithm which only considers the network structure or the content of attributes.
【作者单位】： 武汉理工大学计算机科学与技术学院;武汉理工大学交通物联网技术湖北省重点实验室;武汉理工大学内河航运技术湖北省重点实验室;华中科技大学电子信息与通信学院;
【基金】：国家自然科学基金资助项目(61572219,61502192,61671216,61471408,51479157,51679182) 中央高校基本科研业务费专项资金资助项目(WUT:2016Ⅲ028) 内河航运技术湖北省重点实验室基金资助项目(NHHY2015005)
【分类号】：O157.5;TP301.6
本文编号：2463724
[Abstract]:In this paper, a constrained anomaly detection algorithm for heterogeneous information networks with rich attributes is proposed, aiming at the distortion and difficulty of finding outliers in heterogeneous information networks considering only the network structure. By modeling the information-rich interactive data into an attribute-rich heterogeneous information network, using attribute element paths to specify the attributes and subspaces of interest to the user, the anomaly degree of the nodes is evaluated by integrating the network structure and attribute content. The framework of constrained anomaly detection algorithm is presented. Experiments are carried out on Arxiv real data sets. Constraints on authors, papers, titles and abstracts are specified with attribute element paths, and outliers from high to low in node list and constraint domain outliers are outputted for multiple queries. The results show that the average accuracy of the benchmark algorithm is 12.95% higher than that of the benchmark algorithm which only considers the network structure or the content of attributes.
【作者单位】： 武汉理工大学计算机科学与技术学院;武汉理工大学交通物联网技术湖北省重点实验室;武汉理工大学内河航运技术湖北省重点实验室;华中科技大学电子信息与通信学院;
【基金】：国家自然科学基金资助项目(61572219,61502192,61671216,61471408,51479157,51679182) 中央高校基本科研业务费专项资金资助项目(WUT:2016Ⅲ028) 内河航运技术湖北省重点实验室基金资助项目(NHHY2015005)
【分类号】：O157.5;TP301.6
【相似文献】
相关期刊论文 前3条
1 邓树勇,钟延;试论统计信息网络的建立与管理[J];统计与预测;1998年05期
2 谢心兰;巩固农村基层统计信息网络发挥统计的整体功能作用[J];统计与预测;2003年02期
3 石中元;警惕:信息网络对社会的负影响[J];当代世界;1997年09期
相关硕士学位论文 前2条
1 刘杨杨;信息网络中流问题的灵敏度分析[D];国防科学技术大学;2014年
2 朱天晨;极性信息网络节点相关性度量问题研究[D];山东大学;2017年
，本文编号：2463724
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2463724.html
