A graph anomalies detection method based on graph similarity

作者：Li, Tao; Xiao, Nanfeng

来源：Journal of Xi'an Jiaotong University, 2014, 48(8): 67-72 and 79.

DOI：10.7652/xjtuxb201408012

• Accuracy problems
• Anomalies detections
• Frequent subgraphs
• Graph
• Local perspective
• Outlier Detection
• Similarity
• Steady-state distributions

An anomalies detection method based on considering both the global and the local perspectives is proposed to solve the low accuracy problem in traditional anomalies detection, and is successfully applied to outlier detection in transaction graph data sets. The method evaluates the similarity between any two graphs based on maximum common frequent subgraph, and then cuts out the similarity matrix based on common neighbor. The round-trip distance for a data node is calculated and is used as its anomalies score, that makes up the defect of traditional outlier detection methods based on the steady-state distribution and random walk. Experiments in real datasets show that the performance of the proposed method is better than the performance of the method based on subdue. The precision, the recall rate and the false alarm rate are improved by about 10%.

• 出版日期2014-8-10
• 单位华南理工大学; 华南农业大学