WebDS / README.md

yamhm

Update README.md

a7a631b verified 9 months ago

preview code

raw

history blame contribute delete

5 kB

metadata

datasets:
  - name: WebDS
    license: mit
    task_categories:
      - web-automation
      - multi-hop-reasoning
    language:
      - en
    size_categories:
      - 10K<n<100K
    source_datasets: []
    pretty_name: WebDS

WebDS: A Benchmark for Web-based Data Science

WebDS is the first end-to-end benchmark designed for evaluating agents on real-world web-based data science workflows. It contains 870 tasks across 29 containerized websites spanning 10 domains, including economics, health, climate, and scientific research.

Agents are tested on:

Multi-hop web navigation
Structured and unstructured data processing
Tool usage (e.g., Python scripts, visualization tools)
Downstream task completion (e.g., reports, Reddit posts)

Tasks reflect realistic data science scenarios, such as acquiring data from government portals, comparing datasets across sites, and synthesizing insights in report-ready formats.

📦 Contents

This repository includes:

tasks/: JSON files for all 870 benchmark tasks, with metadata and intents
websites/: Dockerized replicas of 29 benchmark sites for reproducibility
webds_experiments/: Code for running LLM-based agents and collecting evaluation metrics

🌍 Hosted Demo (Docker)

You can try a live version of the benchmark via: http://ec2-18-220-211-153.us-east-2.compute.amazonaws.com:3333

This is useful for previewing the benchmark environment or debugging agent behavior before running large-scale evaluations.

📊 Evaluation

WebDS supports both:

Automatic scoring via reference ground truths (for QA-type tasks)
LLM-as-a-Judge scoring with 1–5 granular feedback and error attribution (for open-ended tasks)

📜 Citation

If you use WebDS in your research, please cite:

@inproceedings{yam2025webds,
  title = {WebDS: An End-to-End Benchmark for Web-based Data Science},
  author = {Yam, Hong Meng and Hsu, Ethan and Bouissou, Ines and John, Aaron Murali and Thota, Raj and Koe, Josh and Putta, Vivek Sarath and Dharesan, G K and Spangher, Alexander and Murty, Shikhar and Huang, Tenghao and Manning, Christopher D.},
  booktitle = {ArXiV},
  year = {2025}
}

Setup of docker file

This file host the instructions for our Docker image.

Docker image

Download the image webbenchdocker.tar.gz from the following link: https://drive.google.com/drive/folders/1LnBfeUqwDm6kiUxDC-vF7vADsWdaAqHp

docker load --input webbenchdocker.tar.gz
docker run --name webbench -p 3333:80 -d webbenchdocker

Shopping Website (OneStopShop) from Webarena

Download the image tar from the following mirrors:

docker load --input shopping_final_0712.tar
docker run --name shopping -p 7770:80 -d shopping_final_0712
# wait ~1 min to wait all services to start

docker exec shopping /var/www/magento2/bin/magento setup:store-config:set --base-url="http://<your-server-hostname>:7770" # no trailing slash
docker exec shopping mysql -u magentouser -pMyPassword magentodb -e  'UPDATE core_config_data SET value="http://<your-server-hostname>:7770/" WHERE path = "web/secure/base_url";'
docker exec shopping /var/www/magento2/bin/magento cache:flush

Now you can visit http://<your-server-hostname>:7770.

E-commerce Content Management System (CMS) from webarena

Download the image tar from the following mirrors:

docker load --input shopping_admin_final_0719.tar
docker run --name shopping_admin -p 7780:80 -d shopping_admin_final_0719
# wait ~1 min to wait all services to start

docker exec shopping_admin /var/www/magento2/bin/magento setup:store-config:set --base-url="http://<your-server-hostname>:7780" # no trailing slash
docker exec shopping_admin mysql -u magentouser -pMyPassword magentodb -e  'UPDATE core_config_data SET value="http://<your-server-hostname>:7780/" WHERE path = "web/secure/base_url";'
docker exec shopping_admin /var/www/magento2/bin/magento cache:flush

Now you can visit http://<your-server-hostname>:7780/admin.

Social Forum Website (Reddit)

Download the image tar from the following mirrors:

docker load --input postmill-populated-exposed-withimg.tar
docker run --name forum -p 4444:80 -d postmill-populated-exposed-withimg

Now you can visit http://<your-server-hostname>:4444/.