FediData: A Comprehensive Multi-Modal Fediverse Dataset from Mastodon

FediData, the first open multi-modal dataset collected from Mastodon, which is dedicated to providing realistic and reliable data support for social behavior modeling, multi-modal learning, and research on user interaction mechanisms.

📥 Download Dataset

Citation

If you use FediData in a scientific publication, we kindly request that you cite the following paper:

  @inproceedings{gao2025fedidata,
      title={{FediData: A Comprehensive Multi-Modal Fediverse Dataset from Mastodon}},
      author={Min Gao and Haoran Du and Wen Wen and Qiang Duan and Xin Wang and Yang Chen},
      year={2025},
      booktitle={Proceedings of the 34th ACM International Conference on Information and Knowledge Management (CIKM’25)}
  }

🚀 Quick Start

Prerequisites

• Python 3.8+
• Required packages (see individual module requirements)
• API keys for OpenAI/Qwen models (for classification tasks)

Installation

# Clone the repository
git clone https://github.com/mgao97/FediData.git
cd FediData
pip install requirements.txt

for example: in the data collection module, pip install -r data_collection/userprofile_ugc_download/requirements.txt

📁 Repository Structure

FediData/
├── 📂 data_collection/           # Data collection tools
│   ├── 📂 image_download/       # Image downloading utilities
│   └── 📂 userprofile_ugc_download/  # User and post data collection
├── 📂 bot_detection/            # Social bot detection models
├── 📂 image_category_classification/  # Image classification tools
├── 📂 topic_emotion/            # Topic and emotion analysis
├── 📂 dataset/                  # Raw and processed data
└── README.md                   # Project overview and usage guide

🛠️ Usage Guide

This repository contains multiple modules for different aspects of Mastodon data processing. Each module has its own detailed README with specific usage instructions:

Data Collection

• User Profile & UGC Collection: Complete pipeline for collecting user profiles, social networks, and posts from Mastodon instances
• Image Download: Concurrent image downloader for extracting images from collected posts
• You might either collect the data using our provided code or directly download the anonymized dataset from Zenodo and extract it into the dataset folder.

Data Analysis

• Topic & Emotion Analysis: Topic classification and sentiment analysis of posts
• Bot Detection: Social bot detection using multiple machine learning models
• Image Classification: Automated image categorization using vision-language models

💡 Tip: Each module's README contains detailed prerequisites, configuration steps, and usage examples.

Analysis Details

Topic & Emotion Analysis

• Topic Classification: Automated topic categorization using LLMs
• Emotion Analysis: Sentiment and emotion detection in posts
• Visualization: Comprehensive charts and comparative analysis

Bot Detection Models

Model	Description	Type
BECE	Bot detection using embedding and classification	Deep Learning
BotRGCN	Relational Graph Convolutional Network	Graph Neural Network
SGBot	Statistical and graph-based features	Random Forest

Image Classification

• Qwen 2.5 VL-32B Instruct: Vision-language model for image categorization
• Supports batch processing with configurable thread pools
• Automatic retry and error handling