ProvFusion — Artifacts & Data

Official artifacts for "Beyond Nodes vs. Edges: A Multi-View Fusion Framework for Provenance-Based Intrusion Detection" (Fan Yang, Binyan Xu, Di Tang, Kehuan Zhang).

Code & documentation: https://github.com/Joney-Yf/ProvFusion_Private

This dataset repo contains everything needed to reproduce the paper's node-level detection results on all nine benchmarks — six DARPA Transparent Computing (TC) datasets (E3/E5: CADETS, THEIA, CLEARSCOPE) and three DARPA OpTC hosts (H201/H051/H501).

Repository layout

release_assets/
├── checkpoints/          # saved middle results; each re-evaluates to exactly the
│                         # paper numbers (see GitHub README for the mapping)
├── ground_truth/         # per-dataset refined ground-truth labels + attack maps
├── regenerated_data/     # merged training inputs regenerated end-to-end from raw
│                         # DARPA logs with our data_preparation pipeline (Option D)
├── *.pt                  # merged training inputs on the original data (Option A/B)
└── MANIFEST.md5          # integrity checksums for every released file

raw_data/{DATASET}/       # per-graph embeddings (train/ val/ test/ *.TemporalData.simple)
                          # for all nine datasets; lets you rebuild the merged .pt
                          # yourself (Option C)

Download

# everything (~310 GB)
hf download Joney233/ProvFusion --repo-type dataset

# one dataset's raw_data only
hf download Joney233/ProvFusion --repo-type dataset --include "raw_data/CADETS_E3/*"

# the release bundle only (checkpoints + merged data + labels, ~41 GB)
hf download Joney233/ProvFusion --repo-type dataset --include "release_assets/*"

Verify integrity afterwards with md5sum -c release_assets/MANIFEST.md5.

Reproducing the paper numbers

Each checkpoint under release_assets/checkpoints/ re-evaluates to exactly the reported TP/FP using the evaluation entry point in the GitHub repo (try_different_threshold.method_12_with_different_normalization, percentile normalization). See the GitHub README for the full checkpoint ↔ result mapping, hyperparameters, and the four reproduction entry points (A: evaluate checkpoint / B: retrain from merged data / C: rebuild from embeddings / D: full pipeline from raw DARPA logs).

Citation

@inproceedings{yang2026provfusion,
  title   = {Beyond Nodes vs. Edges: A Multi-View Fusion Framework for
             Provenance-Based Intrusion Detection},
  author  = {Yang, Fan and Xu, Binyan and Tang, Di and Zhang, Kehuan},
  year    = {2026}
}