Year: 2025
Type: conference-paper
Abstract: Graph anomaly detection (GAD) detects anomalous nodes in real-world networks by capturing topological and attributive information. Although a few of benchmark datasets are publicly available, there is a lack of dynamic heterogeneous graph datasets for advanced GAD research. To address this issue, this paper presents FiGraph, a real-world dynamic heterogeneous graph with ground-truth labels for financial anomaly detection. It consists of nine graph snapshots from 2014 to 2022 and comprises 730,408 nodes and 1,040,997 edges. There are five types of nodes and four types of edges. Only partial nodes (target nodes) need to be identified, and these nodes have multimodal attributes that incorporate tabular data and textual input. The target nodes that correspond to the same entity in different snapshots may have different labels. The remaining nodes do not need to be categorized, serving as background nodes without attributes. In addition, multiple relations can exist simultaneously between the same pair of nodes. For example, two companies may share investment and supply chain relations, while a company and a human may share both investment and related-party transaction relations. These characteristics make FiGraph more realistic and complex than existing GAD datasets, encouraging the development of more effective GAD models. This paper details the construction and properties of FiGraph and discusses promising use cases. The dataset is publicly available at: https://github.com/XiaoguangWang23/FiGraph.
Authors: Xiaoguang Wang, Chenxu Wang, Huanlong Liu, Mengqin Wang, Tao Qin, Pinghui Wang
Institution: Xi'an Jiaotong University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411549709