大规模图异常检测模型数据集

大规模图异常检测模型数据集，即DGraph数据集主要面向金融风控研究、反金融欺诈需求建设，该数据集由浙江大学杨洋教授科研团队与信也科技联合发布，源自真实的金融场景，包含随时间演变的交互式对象、事件和标签。 ​DGraphFin 是一个有向无权的动态图，包含超过 370 万个节点和 430 万条动态边，节点表示金融借贷用户，边表示用户之间的紧急联系人关系。​每条边都关联有时间戳（范围为 1 到 821）和紧急联系人类型（范围为 0 到 11）。 节点特征为 20 维向量，标签包含四类，其中类 1 代表欺诈用户，类 0 代表正常用户，类 2 和类 3 为背景用户，无需预测其标签。 ​该数据集可用于节点分类、链接预测和图分类等任务，特别适用于金融领域的异常检测研究。数据量341.86MB。

The DGraph dataset, a large-scale dataset for graph anomaly detection models, is constructed for financial risk management research and anti-financial fraud needs. It is jointly released by the research team led by Professor Yang Yang from Zhejiang University and FinVolution, and originates from real financial scenarios, containing time-evolving interactive objects, events and labels.

DGraphFin is a directed, unweighted dynamic graph with over 3.7 million nodes and 4.3 million dynamic edges. Nodes represent financial loan users, and edges represent emergency contact relationships between users. Each edge is associated with a timestamp (ranging from 1 to 821) and an emergency contact type (ranging from 0 to 11).

Node features are 20-dimensional vectors. The labels include four categories: category 1 represents fraudulent users, category 0 represents normal users, while categories 2 and 3 are background users whose labels do not require prediction.

This dataset can be applied to tasks such as node classification, link prediction and graph classification, and is particularly suitable for anomaly detection research in the financial domain. The total size of the dataset is 341.86 MB.

搜集汇总

背景与挑战

背景概述

大规模图异常检测模型数据集（DGraph）是一个面向金融风控和反欺诈研究的动态图数据集，由浙江大学杨洋教授科研团队与信也科技联合发布，源自真实金融场景。它包含超过370万个节点和430万条动态边，节点表示金融借贷用户，边表示紧急联系人关系，并带有时间戳和类型，适用于节点分类、链接预测和图分类等异常检测任务。