All data of MARCOVul

Data files include two zip files: dataset.zip and AllData.zip. Where AllData.zip includes everything in dataset.zip.

dataset.zip

dataset.zip includes only raw data and derived data, excluding intermediate data and model data.

AllData.zip

The AllData folder structure is the same as the code folder structure.

The dataset folder contains:

DiverseVul: The original dataset. Among them, diversevul_20230702.json is the obtained raw data, while train.jsonl, valid.jsonl, and test.jsonl are the randomly split training, validation, and testing sets.

DiverseVul-Derived: The dataset after data derivation. Among them, allData.jsonl contains all the data with verified functions before data derivation, while train.jsonl, valid.jsonl, and test.jsonl are the training, validation, and testing sets after data derivation.

The joern/data folder contains all generated intermediate structure data:

DiverseVul-Derived: Data after intermediate structure generation.

DiverseVul-Derived-enhance: Data after data augmentation.

DiverseVul-Derived-simplify: Data after data simplification.

DiverseVul-Derived-simplify-enhance: Data after both data simplification and data augmentation.

The MultiFCode/LanguageModel folder contains:

pre-model: The raw language models.

saved_models: The language models fine-tuned on the training set.