MarcDamie/Fedivertex

Introduction

Social network graphs are central to graph learning research, serving as standard benchmarks for algorithm evaluation. However, existing datasets focus mainly on mainstream social media platforms whose structures are shaped notably by algorithmic recommendations. This raises an important question: would alternative, decentralized social networks exhibit different properties?

We address this by studying the Fediverse; a collection of decentralized social networks (such as Mastodon and Lemmy). These platforms differ fundamentally from for-profit social media, notably in decentralization and absence of recommendation algorithms, which may yield distinct graph structures.

We introduce Fedivertex, a dataset of over 400 graphs from seven decentralized networks, collected weekly over more than a year. The dataset, released with a companion Python package to facilitate its use, supports research on temporal and structural aspects of decentralized social networks.

Citation

@inproceedings{10.1145/3774904.3792868,
  author = {Damie, Marc and Cyffers, Edwige},
  title = {Fedivertex: a Graph Dataset based on Decentralized Social Media},
  year = {2026},
  isbn = {9798400723070},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {https://doi.org/10.1145/3774904.3792868},
  doi = {10.1145/3774904.3792868},
  booktitle = {Proceedings of the ACM Web Conference 2026},
  pages = {8393–8396},
  numpages = {4},
  location = {United Arab Emirates},
  series = {WWW '26}
}

Python Package

We implemented a simple Python API to interact easily with the dataset: https://pypi.org/project/fedivertex/

pip3 install fedivertex

This package automatically downloads the dataset and generate NetworkX graphs.

from fedivertex import GraphLoader

loader.list_graph_types("mastodon")
# List available graphs for a given software, here federation and active_user

G = loader.get_graph(software = "mastodon", graph_type = "active_user", index = 0, only_largest_component = True)
# G contains the Networkx graph of the giant component of the active users graph at the 1st date of collection

Available graphs

The dataset contains graphs crawled on a daily basis on 7 social networks from the Fediverse. Each graph quantifies/characterizes the interaction differently depending on the information provided by the public API of these networks.

We present briefly the graph below (NB: the term "instance" refers to servers on the Fediverse):

• [Bookwyrm/Friendica/Lemmy/Misskey/Pleroma] "federation" graphs: If two instances know each other they are connected in this graph. The federation graph then corresponds to the undirected communication graph between instances.
• Peertube "follow" graphs: On Peertube, an instance X can follow an instance Y to let its users see all the videos posted on Y. This graph is a directed graph with edges of weight 1 for following.
• Lemmy "federation with blocks" graphs: This graph completes the federation graph with negative edges when an instance X blocks instance Y. The graph is directed.
• Lemmy "cross-instance" graphs: two instances are connected as soon as there exists a pair of users who published a message in the same thread, but possibly on a third instance. This is an undirected graph, less sparse than "intra-instance".
• Lemmy "intra-instance" graphs: the instance X is linked to Y if an user of X has published a message on instance Y. This graph is directed and very sparse.
• [Mastodon/Misskey/Pleroma] "active users" graphs: For each instance, we consider the set of the 10K most recently active users. Then, for each user of an instance X, we consider the list of the users they follow, and add 1 to the edge from X to Y where Y is the instance the followed users. The weight of the edge from X to Y thus encodes how much the content seen on instance X is generated in instance Y. Note that this graph thus contains self loops.

These graphs provide diverse perspectives on the Fediverse as they capture more or less subtle phenomenon. For example, "federation" graphs are dense, while "intra-instance" graphs are sparse. We have performed a detailed exploratory data analysis in this notebook.

Reduced dataset

The complete dataset requires 20 GB of storage, mostly due to large graphs based on the Mastodon federation. We released a reduced version excluding these large graphs, resulting in a 2GB dataset: https://huggingface.co/datasets/MarcDamie/Fedivertex-Reduced