A Graph Encoder-Decoder Network for Unsupervised Anomaly Detection

Detecting anomalies in a graph typically involves identifying nodes that deviate significantly from the normal behavior A key component of many graph neural networks (GNNs) is the of the graph, either in terms of their structural characteristics pooling operation, which seeks to reduce the size of a graph and/or their feature attributes [3]. However, graphs can often be while preserving important structural information. However, very large and complex, making it challenging to identify such most existing graph pooling strategies rely on an assignment anomalies. To address this problem, graph pooling can be used matrix obtained by employing a GNN layer, which is characterized to reduce the size of the graph while preserving its important by trainable parameters, often leading to significant computational structural features [4-6]. The aim is to produce a coarse representation complexity and a lack of interpretability in the pooling of the graph structure by summarizing the information process. In this paper, we propose an unsupervised graph contained in the nodes of the graph into a fixed-size vector or encoder-decoder model to detect abnormal nodes from graphs matrix while preserving the salient features of the graph. By by learning an anomaly scoring function to rank nodes based producing a coarse representation of the graph structure, graph on their degree of abnormality. In the encoding stage, we design pooling can help abstract away irrelevant or noisy details, and a novel pooling mechanism, named LCPool, which leverages focus on the most important structural properties of the graph.