Vulnerability Dataset

This repository contains a dataset for vulnerability detection used in our research study:
A Zero-Shot Framework for Cross-Project Vulnerability Detection in Source Code.

📂 Dataset Overview

The dataset is derived from publicly available datasets used in prior research on vulnerability detection:

• Devign: Effective Vulnerability Identification by Learning Comprehensive Program Semantics via Graph Neural Networks (DOI: 10.48550/arXiv.1909.03496)
• REVEAL: Deep Learning based Vulnerability Detection: Are We There Yet? (DOI: 10.48550/arXiv.2009.07235)

To facilitate ease of use, we have preprocessed and reformatted the datasets into a single combined.pkl file containing labeled source code functions.

🔹 File

File Name	Description
combined.pkl	A dictionary containing vulnerability-labeled functions from multiple projects (FFmpeg, Chrome, Debian, Qemu).

🔹 Structure

combined.pkl contains a dictionary where:

• combined_data["FFmpeg"]
• combined_data["Chrome"]
• combined_data["Debian"]
• combined_data["Qemu"]

Each dataset consists of source code functions and their binary labels indicating whether they are vulnerable (1) or non-vulnerable (0).

📖 Usage

To load the dataset in Python:

import pickle

with open("combined.pkl", "rb") as f:
    combined_data = pickle.load(f)

# Access datasets
ffmpeg_data = combined_data["FFmpeg"]
chrome_data = combined_data["Chrome"]
debian_data = combined_data["Debian"]
qemu_data = combined_data["Qemu"]

print(ffmpeg_data)