Hugging Face's logo

Datasets:
goker
/
comp-serp-data-v2

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
100K - 1M
Tags:
serp
search-engines
information-retrieval
web-scraping
reproducibility
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
comp-serp-data-v2 / README.md
goker's picture
goker
Update README.md
3b8d3ac verified
preview code
|
raw
history blame contribute delete
2.65 kB
metadata 
license: mit
language:
  - en
pretty_name: Comprehensive SERP Data (v2)
tags:
  - serp
  - search-engines
  - information-retrieval
  - web-scraping
  - reproducibility

Comprehensive SERP Data (v2)

A reproducible dataset for studying how news-derived search terms propagate through multiple search engines, and how technical/content/accessibility signals vary across ranked results.

This repository contains:

  • the query/keyword seeds,
  • SERP indexing outputs,
  • the derived feature dataset used for analysis,
  • and documentation reports describing acceptance, outliers, and dataset health.

What’s inside (high level)

Core inputs (analysis seeds)

  • keywords.csv: news headline–derived search terms and related metadata (title, url, section, pub_date, word_count, text_saved_path).
    (Report: ~3.3k rows; date range spans 2025-10 to 2026-01.)

Index & derived data products

  • index.parquet: indexed SERP results (record-level index; includes engine, rank, url, collection metadata). Check the index_stats.md for more details.
  • dataset-20260222_141303.parquet: final derived dataset used in analysis.

The dataset was generated from index.parquet with an acceptance configuration and optional outlier handling; see reports/ for full details.

Dataset snapshot (this release)

  • Final dataset file: dataset-20260222_141303.parquet
  • Generated: 2026-02-22
  • Total records (final): 86,563
  • Engines: Google, Brave, Mojeek
  • Rank range: 1–20

Acceptance / filtering overview

The final dataset is produced by applying an acceptance policy (e.g., status=ok, HTTP 200, minimum content, required similarity fields, and required runtime/accessibility metrics). See:

Outliers & consistency

Outlier detection and extraction consistency checks are summarized in:

Repository structure (recommended) 

├── data/
│   ├── keywords.csv
│   ├── index.parquet
│   └── dataset-20260222_141303.parquet
├── raw/        
│   ├── scraping/
│   ├── serp/
└── reports/
    ├── dataset-20260222_141303.md
    ├── extraction_*.md
    ├── index_stats.md
    └── outlier_report.md

How to use

Our collection, extraction, and analysis pipeline is available in the serp-profiler-kit repository.

License

This project is licensed under the MIT License - see the LICENSE file for details.