README.md

metadata

dataset_info:
  features:
    - name: project
      dtype: string
    - name: commit_id
      dtype: string
    - name: CVE ID
      dtype: string
    - name: CWE ID
      dtype: string
    - name: func
      dtype: string
    - name: vul
      dtype: int8
  splits:
    - name: train
      num_bytes: 299777443
      num_examples: 239822
    - name: test
      num_bytes: 63350463
      num_examples: 51390
    - name: validation
      num_bytes: 63678823
      num_examples: 51392
  download_size: 190857204
  dataset_size: 426806729
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
license: mit
task_categories:
  - text-classification
  - feature-extraction
tags:
  - Code
  - Vulnerability

Merged BigVul and PrimeVul Dataset

Dataset ID: mahdin70/merged_bigvul_primevul

This dataset is a merged and preprocessed combination of the BigVul (bstee615/bigvul) and PrimeVul (colin/PrimeVul, "default" configuration) datasets, designed for vulnerability analysis and machine learning tasks. The preprocessing ensures consistency in column names, data types, and formats, making it suitable for fine-tuning models.

Dataset Overview

The dataset integrates vulnerability data from two sources:

• BigVul: A dataset of real-world vulnerabilities from open-source C/C++ projects.
  • Paper: (https://doi.org/10.1145/3379597.3387501)
  • Repository: (https://github.com/ZeoVan/MSR_20_Code_vulnerability_CSV_Dataset)
• PrimeVul: A vulnerability dataset with additional project-specific details.
  • Paper: (https://doi.org/10.48550/arXiv.2403.18624)
  • Repository:(https://github.com/DLVulDet/PrimeVul)

The merged dataset retains key information about projects, commits, functions, and vulnerabilities, standardized for consistency.

Columns

The dataset contains the following columns:

• project: String - The name of the project (e.g., "qemu", "linux-2.6").
• commit_id: String - Unique identifier of the commit associated with the function.
• func: String - The source code of the function before fixing (from func_before in BigVul).
• vul: Int8 - Vulnerability label (1 = vulnerable, 0 = not vulnerable).
• CVE ID: String - Common Vulnerabilities and Exposures identifier (e.g., CVE-2007-1320), or NOT_APPLICABLE if vul = 0.
• CWE ID: String - Common Weakness Enumeration identifier (e.g., CWE-20), or NOT_APPLICABLE if vul = 0.

Splits

• Train: Combined training data from BigVul and PrimeVul.
• Test: Combined testing data from BigVul and PrimeVul.
• Validation: Combined validation data from BigVul and PrimeVul.

Preprocessing Steps

The dataset was preprocessed to ensure consistency and quality:

BigVul Preprocessing

• Source Columns:
  • project, commit_id, CVE ID, CWE ID, func_before, vul.
• Transformations:
  • Renamed func_before to func.
  • Kept CWE ID in its original format (CWE-XXX).
  • Converted vul to int8.

PrimeVul Preprocessing

• Source Columns:
  • project, commit_id, cve, cwe, func, target.
• Transformations:
  • Renamed cve to CVE ID, cwe to CWE ID, target to vul.
  • Standardized CWE ID by removing brackets from list format (e.g., ["CWE-XXX"] → CWE-XXX), taking the first element if multiple CWEs exist.
  • Converted vul from int64 to int8.

Merging and Final Preprocessing

• Merging: Concatenated preprocessed BigVul and PrimeVul data for each split (train, test, validation).
• Final Steps:
  • Removed rows with any null values.
  • Removed duplicates based on the func column.
  • For rows where vul = 0, replaced CVE ID and CWE ID with "NOT_APPLICABLE.

Dataset Statistics

Below are the analysis results for the final merged dataset:

Train Split

• Number of rows: 239,822
• Unique CWE IDs (excluding NOT_APPLICABLE): 127
• Unique commit IDs: 7,559
• Vulnerable functions (vul = 1): 9,037

Test Split

• Number of rows: 51,390
• Unique CWE IDs (excluding NOT_APPLICABLE): 87
• Unique commit IDs: 6,032
• Vulnerable functions (vul = 1): 1,911

Validation Split

• Number of rows: 51,392
• Unique CWE IDs (excluding NOT_APPLICABLE): 91
• Unique commit IDs: 6,059
• Vulnerable functions (vul = 1): 1,933

Usage

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("mahdin70/merged_bigvul_primevul")