Farcaster Threads Dataset

Overview

This dataset contains high-quality thread data from Farcaster's entire history, featuring 512-dimensional embeddings generated using VoyageAI (float32) on formatted thread text. The dataset includes comprehensive thread metadata, engagement metrics, and vector embeddings suitable for semantic search, recommendation systems, and content analysis.

Dataset Details

• Total Records: ~18,147,313 threads
• Data Cutoff: 2025-06-25 (no threads newer than this date)
• Quality Filter: Non-spam content only (spam label = 2), low-effort replies and threads removed.
• Embedding Model: VoyageAI 512-dimensional float32 vectors
• Repository: shoni/farcaster
• Data Location: /threads folder

Schema

Column	Type	Description
hash	bytes	Unique thread identifier (primary key)
fids	list[int]	FIDs of reply authors, ordered by reaction count (most reacted first)
reactions	int	Total reaction count from non-spam users
author_fid	int	FID of the original thread author
timestamp	timestamp	Thread creation date
claimed_at	timestamp	When thread was claimed by processing worker
blob	string	Formatted thread text including replies and author information
blob_embedding	list[float32]	512-dimensional VoyageAI embeddings of the blob text
blob_timestamp	timestamp	Timestamp of when the blob text was generated
blob_embedding_binary	bytes	Binary representation of the embeddings (512 bits)

Data Quality

• Spam Filtering: Only includes threads with spam label = 2 (verified non-spam)
• Engagement Metrics: Reaction counts filtered to exclude spam users
• Historical Coverage: Complete Farcaster thread history up to cutoff date
• Reply Ordering: Reply authors (fids) sorted by engagement for relevance

Use Cases

• Semantic search over Farcaster content
• Thread recommendation systems
• Content similarity analysis
• Social network analysis
• Engagement prediction modeling
• Community detection and analysis

Loading the Dataset

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("shoni/farcaster", data_dir="threads")

# Access embeddings
embeddings = dataset['train']['blob_embedding']

# Access formatted thread text
threads = dataset['train']['blob']

Citation

If you use this dataset in your research, please cite:

@dataset{farcaster_open_social_dataset_2025,
  title={Farcaster Open Social Dataset},
  author={shoni},
  year={2025},
  url={https://huggingface.co/datasets/shoni/farcaster}
}

License

CC-BY-4.0

Example Thread

Note: The actual blob_embedding content in the dataset uses private formatting techniques that slightly differ from the displayed blob text in the dataset. The displayed blob text is a human-readable format, replies are ordered from most to least popular (as whole branches):

@keremgurel: Beta launch is set for April 24th - exactly 3 months after the project kickoff. We’re building the first truly intuitive no-code website builder powered by blockchain, on @base https://imagedelivery.net/BXluQx4ige9GuW0Ia56BHw/5a93839e-f6be-4c04-b66a-4c2783328600/original https://stream.warpcast.com/v1/video/0195eca0-eccd-3042-b676-2f1bbd179695.m3u8
~1: @compusophy: cool!! though no need to use the word first here
~2: @keremgurel: Thanks! Why so? To my knowledge there aren’t any other onchain website builders out there
~1: @damag.eth: where can i sign up 👀
~2: @keremgurel: https://onchainsite.xyz https://onchainsite.xyz/

Repository

https://huggingface.co/datasets/shoni/farcaster