.png){kind=icon}
.png){kind=icon}
Return
Dual-channel graph-level anomaly detection method based on multi-graph representation learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-024-05852-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph-level anomaly detection plays a crucial role in anomaly identification by comparing and classifying the graph-level features of normal and anomalous graphs. Despite advancements, existing methods often suffer from low detection rates and high false-positive rates when dealing with sparse anomalous data. To address this limitation, we propose a dual-channel graph-level anomaly detection model that utilizes two graph isomorphic networks to separately learn from labeled anomalous data and unlabeled normal data. This model enhances the identification of unlabeled anomalies by learning from both types of data through separate channels. Furthermore, to enable the model to be applicable to complex graph types in graph-level anomaly detection applications, we introduce a novel multi-graph representation learning method that can transform multi-graphs into a simplified graph representation. We have rigorously evaluated the proposed model on 6 public datasets, and the experimental results demonstrate the effectiveness of the model, with significant performance improvements over 9 baseline models.
Keywords:
Graph-level anomaly detection
Graph representation learning
Multi-graph
Anomalous data
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W