soumikmahato/diverse-websearch-3.5k

Diverse WebSearch 3.5k

Diverse WebSearch 3.5K is a small web research dataset containing 3,500 diverse web pages collected from search results. Each row contains a source URL, extracted markdown content, a concise summary, and image URLs found on the page.

Dataset Details

This dataset is intended for learning and experimentation with:

• webpage summarization
• retrieval-augmented generation
• search result understanding
• document cleaning
• synthetic QA generation
• dataset curation pipelines

Columns

Column	Description
url	Source webpage URL
domain	Domain extracted from the URL
markdown	Extracted webpage content in markdown-like text
summary	Summary of the webpage
image_urls	List of image URLs found on the webpage
markdown_length	Character length of the markdown field
summary_length	Character length of the summary field
task_type	Dataset task type, currently webpage_summarization

Suggested Usage

from datasets import load_dataset

ds = load_dataset("soumikmahato/diverse-websearch-3.5k")
print(ds["train"][0])

Example Task Format

Input:

Summarize the following webpage:

<markdown>

Output:

<summary>

Limitations

The dataset is small and intended mainly for educational use. Some webpage extractions may contain boilerplate, navigation text, duplicated content, or incomplete markdown. Users should apply additional filtering before using it for model training.

Data Considerations

The dataset contains URLs and extracted webpage text from public web pages. Before using this dataset commercially or redistributing derived versions, users should review the licensing and terms of the original source websites.

Intended Use

Recommended uses:

educational experiments RAG corpus construction summarization experiments dataset cleaning and labeling practice search/research agent prototyping