This website requires JavaScript.

Unsupervised Graph Outlier Detection: Problem Revisit, New Insight, and Superior Method

Yihong HuangLiping WangFan ZhangXuemin Lin
Oct 2022
A large number of studies on Graph Outlier Detection (GOD) have emerged inrecent years due to its wide applications, in which Unsupervised Node OutlierDetection (UNOD) on attributed networks is an important area. UNOD focuses ondetecting two kinds of typical outliers in graphs: the structural outlier andthe contextual outlier. Most existing works conduct experiments based ondatasets with injected outliers. However, we find that the most widely-usedoutlier injection approach has a serious data leakage issue. By only utilizingsuch data leakage, a simple approach can achieve state-of-the-art performancein detecting outliers. In addition, we observe that most existing algorithmshave a performance drop with varied injection settings. The other major issueis on balanced detection performance between the two types of outliers, whichhas not been considered by existing studies. In this paper, we analyze the cause of the data leakage issue in depth sincethe injection approach is a building block to advance UNOD. Moreover, we devisea novel variance-based model to detect structural outliers, which outperformsexisting algorithms significantly at different injection settings. On top ofthis, we propose a new framework, Variance-based Graph Outlier Detection(VGOD), which combines our variance-based model and attribute reconstructionmodel to detect outliers in a balanced way. Finally, we conduct extensiveexperiments to demonstrate the effectiveness and efficiency of VGOD. Theresults on 5 real-world datasets validate that VGOD achieves not only the bestperformance in detecting outliers but also a balanced detection performancebetween structural and contextual outliers. Our code is available at