PyPI FCG and TPKG Dataset
This dataset contains the final graph artifacts used in the capstone project: Detecting Python-specific Malicious Behaviors in PyPI Packages based on Threat Pattern Knowledge Graph and AST Analysis.
Dataset Summary
- Final metadata: 1,428 PyPI package-version samples
- Valid graph samples: 1,416
- Valid malicious graph samples: 715
- Valid benign graph samples: 701
- Split strategy: temporal train/validation/test split
Included Artifacts
| Path | Description |
|---|---|
dataset_metadata.csv | Final package-version metadata, including package, version, label, timestamp, and source |
fcg_malnet/ | Raw Function Call Graph edgelist files in a MalNet-style folder structure |
tpg_malnet_qualified/ | Final TPKG-qualified behavior-level graph edgelist files |
tpg_node_features_qualified/ | NumPy node feature files for the fixed 11 behavior nodes |
split-info-fcg/ | Temporal train/validation/test split files |
file_csv_data/fcg_edges_qualified.csv | Qualified FCG edge table with resolved API names and confidence values |
Notes
The original extracted package source code is not included in this final public dataset artifact because it may contain real malicious code and can be large. The released files are the graph-level artifacts used for model training, baseline evaluation, size bias analysis, and explainability experiments.
Labels
0: benign1: malicious
Citation
If you use this dataset, please cite the capstone project or refer to the accompanying project report.