Community Distribution Outlier Detection in Heterogeneous Information Networks [chapter]

Manish Gupta, Jing Gao, Jiawei Han
2013 Lecture Notes in Computer Science  
Heterogeneous networks are ubiquitous. For example, bibliographic data, social data, medical records, movie data and many more can be modeled as heterogeneous networks. Rich information associated with multi-typed nodes in heterogeneous networks motivates us to propose a new definition of outliers, which is different from those defined for homogeneous networks. In this paper, we propose the novel concept of Community Distribution Outliers (CDOutliers) for heterogeneous information networks,
more » ... h are defined as objects whose community distribution does not follow any of the popular community distribution patterns. We extract such outliers using a type-aware joint analysis of multiple types of objects. Given community membership matrices for all types of objects, we follow an iterative two-stage approach which performs pattern discovery and outlier detection in a tightly integrated manner. We first propose a novel outlier-aware approach based on joint non-negative matrix factorization to discover popular community distribution patterns for all the object types in a holistic manner, and then detect outliers based on such patterns. Experimental results on both synthetic and real datasets show that the proposed approach is highly effective in discovering interesting community distribution outliers. 2 explain the generation of these community distributions. For example, in bibliographic networks, when grouping authors based on their "research area" distributions, one must use the knowledge of the grouping of "research area" distributions for related conferences and keywords. This is because (1) the community space (research areas) is the same across different object types, and (2) all these objects interact strongly with each other in the network.
doi:10.1007/978-3-642-40988-2_36 fatcat:jzf3ku2emfggjo27ouahyqzxha