未提及数据集具体名称
收藏资源简介:
该数据集是针对文本属性图异常检测(TAGAD)而创建的,包含了8个数据集,其中最大的一个数据集包含超过110万个节点和630万个边。这些数据集旨在促进TAGAD领域的研究,帮助研究人员更好地理解和解决文本属性图中的异常检测问题。数据集的创建过程并未在论文中详细描述,但可以推测是通过收集和整理现实世界中的文本属性图数据,并进行标注和处理得到的。这些数据集可以应用于各种场景,如电子商务网络、引文网络等,旨在帮助研究人员提高文本属性图异常检测的准确性和效率。
This dataset is developed for Text-Attributed Graph Anomaly Detection (TAGAD). It includes 8 sub-datasets, among which the largest one contains over 1.1 million nodes and 6.3 million edges. These datasets are intended to facilitate research in the TAGAD domain, enabling researchers to better understand and resolve anomaly detection problems in text-attributed graphs. The specific creation process of these datasets is not elaborated in the associated paper, but it can be inferred that they are generated by collecting and curating real-world text-attributed graph data, followed by annotation and processing. These datasets can be applied to various scenarios such as e-commerce networks, citation networks, and more, with the goal of helping researchers enhance the accuracy and efficiency of text-attributed graph anomaly detection.
CMUCL数据集概述
1. 数据集简介
- 研究领域:文本属性图异常检测
- 方法:基于多尺度跨模态和单模态对比学习
2. 数据集详情
| 数据集 | 节点数 | 边数 | 平均文档长度 | 属性维度 | 异常数 |
|---|---|---|---|---|---|
| Citeseer | 3,186 | 3,432 | 153.94 | 768 | 128 |
| Pubmed | 19,717 | 90,368 | 256.08 | 768 | 788 |
| History | 41,551 | 369,252 | 228.36 | 768 | 1,662 |
| Photo | 48,362 | 512,933 | 150.25 | 768 | 1,934 |
| Computers | 87,229 | 742,792 | 93.16 | 768 | 3,490 |
| Children | 76,875 | 1,574,664 | 209.12 | 768 | 3,076 |
| ogbn-Arxiv | 169,343 | 1,210,112 | 179.70 | 768 | 6,774 |
| CitationV8 | 1,106,759 | 6,396,265 | 148.77 | 768 | 44,270 |
3. 数据获取
- 下载地址:https://drive.google.com/drive/folders/1Suws6A-v0jBQpKeMphD0CEgxBjK-72AC?usp=sharing
- 存放路径:解压后放置于
./data目录
4. 训练与推理
bash python main_train.py --dataset Citeseer --lr 0.0002 --epoch_num 2 --gamma 0.005 python main_train.py --dataset Pubmed --lr 2e-05 --epoch_num 2 --gamma 0.001 python main_train.py --dataset History --lr 2e-05 --epoch_num 2 --gamma 0.5 python main_train.py --dataset Photo --lr 5e-05 --epoch_num 3 --gamma 0.001 python main_train.py --dataset Computers --lr 2e-05 --epoch_num 3 --gamma 0.01 python main_train.py --dataset Children --lr 5e-05 --epoch_num 2 --gamma 0.5 python main_train.py --dataset Arxiv --lr 1e-05 --epoch_num 2 --gamma 0.01 python main_train.py --dataset CitationV8 --lr 2e-5 --epoch_num 2 --gamma 0.5
5. 环境配置
- Python版本:3.8.13
- 核心依赖库:
- torch==1.12.1
- torch-geometric==2.1.0.post1
- transformers==4.24.0
- 相关图形处理库(torch-cluster/scatter/sparse/spline-conv)

- 1Text-Attributed Graph Anomaly Detection via Multi-Scale Cross- and Uni-Modal Contrastive Learning西安交通大学计算机科学与技术学院, 陕西大数据知识工程省级重点实验室, 西安交通大学远程教育学院, 弗吉尼亚大学 · 2025年



