MathiasExorde's picture
Update README.md
7e73c84 verified
metadata
language:
  - multilingual
license: mit
annotations_creators:
  - machine-generated
language_creators:
  - found
size_categories:
  - 10M<n<100M
source_datasets:
  - original
task_categories:
  - text-classification
  - summarization
  - text-retrieval
pretty_name: Exorde Social Media Dataset December 2024 Week 1
tags:
  - social-media
  - multi-lingual
  - sentiment-analysis
  - emotion-detection
  - text

Multi-Source, Multi-Language Social Media Dataset (1 Week Sample)

This dataset represents a rich, diverse snapshot of global online discourse, collected over a one-week period from December 1, 2024, to December 7, 2024. It comprises 65,542,211 entries from various social media platforms, blogs, and news articles, all precisely timestamped at the moment of posting. This dataset is procuded by Exorde Labs. www.exordelabs.com/. This dataset includes many conversations around the period of CyberMonday, Syria regime collapse and UnitedHealth CEO killing & many more topics. The potential is wide. All items in this dataset are captured publicly, in near real-time, allowing post-deletion & retrospective analyses.

Dataset Highlights

  • Multi-Source: Captures content from a wide range of online platforms

  • Multi-Language: Covers 122 different languages

  • High-Resolution Temporal Data: Each entry is timestamped to the exact moment of posting

  • Rich Metadata: Includes sentiment analysis, emotion detection, and thematic categorization

  • Large Scale: 65,542,211 entries collected over just one week

  • Diverse Content: Social media posts, blog entries, news articles, and more

Dataset Schema

  • date: string (exact timestamp of post)
  • original_text: string
  • url: string
  • author_hash: string (SHA-1 hash for privacy)
  • language: string
  • primary_theme: string
  • english_keywords: string
  • sentiment: double
  • main_emotion: string
  • secondary_themes: list<element: int64>

Attributes description

  • original_text is the exact original text of the item/post, as it was collected. It should match the original content before any deletion/edition.
  • author_hash is a SHA-1 Hash of the author username on a given platform, when provided. Many items have None Author_hash.
  • language is detected by a fasttext-langdetect model. Isocode ISO 639.
  • primary_theme is the output of MoritzLaurer/deberta-v3-xsmall-zeroshot-v1.1-all-33, on on the classes below.
  • secondary_themes are the same theme classes with a mapping:
   1. Economy
   2. Technology
   3. Investing
   4. Business
   5. Cryptocurrency
   6. Social
   7. Politics
   8. Finance
   9. Entertainment
   10. Health
   11. Law
   12. Sports
   13. Science
   14. Environment
   15. People
  • main_emotion is computed from an emotion scoring Language model, fine-tuned on social media data.

  • english_keywords is a powerful attribute, computed from an English translation of the original text. These keywords represent the core content (relevant keywords) of the text. They are produced from KeyBert & statistical algorithms. They should be mostly in English except when translation was faulty, in that case they will be in the original language.

  • Sentiment is computed & aggregated from several models, including deep learning models. It is a value between -1 and 1. -1 being negative, 0 neutral and 1 positive.

Key Statistics

  • Total entries: 65,542,211

  • Date range: 2024-12-01 to 2024-12-07 (included)

  • Unique authors: 9,235,952

  • Languages: 122

  • Primary themes: 16

  • Main emotions: 26

  • Average sentiment: 0.051

  • Most common emotion: Neutral

Top 10 Sources

  1. x.com: 49,893,080

  2. reddit.com: 14,385,495

  3. bsky.app: 7,292,262

  4. youtube.com: 1,778,020

  5. 4channel.org: 233,134

  6. jeuxvideo.com: 61,373

  7. forocoches.com: 55,683

  8. mastodon.social: 54,932

  9. news.ycombinator.com: 32,216

Full source distribution

Top 10 Languages

  1. English (en): 53,882,510

  2. Spanish (es): 5,012,521

  3. Japanese (ja): 3,653,281

  4. Portuguese (pt): 3,548,869

  5. French (fr): 1,487,180

  6. German (de): 1,297,085

  7. Turkish (tr): 850,770

  8. Arabic (ar): 829,992

  9. Italian (it): 720,955

Full language distribution

About Exorde Labs

Exorde Labs is pioneering a novel collective distributed data DePIN (Decentralized Physical Infrastructure Network). Our mission is to produce a representative view of the web, minute by minute. Since our inception in July 2023, we have achieved:

  • Current capacity: Processing up to 4 billion elements annually
  • Growth rate: 20% monthly increase in data volume
  • Coverage: A comprehensive, real-time snapshot of global online discourse
  • More than 10 Million data points are processed daily, half a million per hour in near real-time

This dataset is a small sample of our capabilities, offering researchers and developers a glimpse into the rich, multi-faceted data we collect and analyze.

For more information about our work and services, visit:

Use Cases

This dataset is invaluable for a wide range of applications, including but not limited to:

  • Real-time trend analysis

  • Cross-platform social media research

  • Multi-lingual sentiment analysis

  • Emotion detection across cultures

  • Thematic analysis of global discourse

  • Event detection and tracking

  • Influence mapping and network analysis

Citation Information

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

Exorde Labs. (2024). Multi-Source, Multi-Language Social Media Dataset (1 Week Sample) [Data set]. Exorde Labs. https://www.exordelabs.com/

Acknowledgments

We would like to thank the open-source community for their continued support and feedback. Special thanks to all the platforms and users whose public data has contributed to this dataset.

Massive thanks to the Exorde Network and its data enthusiast community, unique of its kind.

Licensing Information

This dataset is released under the MIT license.

Contact Information

For questions, feedback, or more information about this dataset or Exorde Labs' services, please contact us at:

We are committed to supporting the open-source community by providing high-quality, diverse datasets for cutting-edge research and development. If you find this dataset useful, consider exploring our API for real-time access to our full range of social media data.

Exorde  Labs Logo