🤖 GTArena-UI-Defects

[💻 Github] [📖 ArXiv Paper]

This dataset is for 📄"GUI Testing Arena: A Unified Benchmark for Advancing Autonomous GUI Testing Agent".

📖 Overview

Welcome to GTArena-UI-Defects, a comprehensive dataset specifically designed for benchmarking autonomous GUI testing agents. This dataset offers a unique combination of real-world GUI defects, applications with injected defects, and synthetic defect datasets, providing a well-rounded and challenging testbed for researchers and practitioners in the field of automated GUI testing.

Key Features:

• 📊 Diverse Data Sources: Real-world applications, defect-injected apps, and synthetic datasets ensure comprehensive coverage.

• 👾 Controlled Defects: Injected and synthetic defects provide consistency and reproducibility for experimentation.

• 🌐 Real-World Challenges: Naturally occurring defects from real applications ensure authenticity.

🗂️ Dataset Structure

1️⃣ Real-world Applications with GUI Defects • Mining releases & Issue Trackers on Github: Projects are filtered based on releases and issues to identify real occurring defects. • Comprehensive Profiling: Includes defect types, issue severity, and version history. • Validation Process: Reported issues are carefully reproduced to ensure relevance and authenticity.

2️⃣ Applications with Injected Defects • Defects introduced via source code modification. • Controlled Environment: Predefined GUI defects ensure consistent testing scenarios. • Diverse Fault Scenarios: Injection across different areas to simulate real-world complexities.

3️⃣ Synthetic Defect Datasets • Simulated GUI defects created by transforming screenshots of stable applications. Our GUI defect simulation tool is available on here • Designed to overcome the lack of early-stage defects in proprietary commercial apps. • Wide Applicability: Generates defect data even for mature, stable applications.

📊 Data Distribution

Data of Applications	GUI Display	GUI Interaction
Real-World	53
Artificial Inject	79	26
AitW with Defects	6610	1871
Close-Source	1148	399
Open-Source	590	257

🔍 Data Usage

To load and explore the dataset:

from datasets import load_dataset

real_world_data = load_dataset("songjah/GTArena-UI-Defects", split="real_world")
injected_data = load_dataset("songjah/GTArena-UI-Defects", split="injected")
synthetic_data = load_dataset("songjah/GTArena-UI-Defects", split="synthetic")

🌟 Citation

If you use this work in your research or applications, please cite it as follows:

@misc{zhao2024guitestingarenaunified,
      title={GUI Testing Arena: A Unified Benchmark for Advancing Autonomous GUI Testing Agent}, 
      author={Kangjia Zhao and Jiahui Song and Leigang Sha and Haozhan Shen and Zhi Chen and Tiancheng Zhao and Xiubo Liang and Jianwei Yin},
      year={2024},
      eprint={2412.18426},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2412.18426}, 
}