Lesezeichen(i) We introduce Fedivertex, a large and diverse graph dataset based on the Fediverse. More precisely, our dataset encompasses seven Fediverse platforms, resulting in 182 graphs: 13 different graphs each with a sequence of 14 snapshots obtained through weekly web crawls over a period of three months.
(ii) We provide a Python package, fedivertex, available through PyPI, to easily access and use our dataset. The package includes built-in preprocessing tools to download and prepare the graphs for machine learning tasks. We demonstrate its usefulness by benchmarking several existing decentralized learning algorithms.
(iii) We formalize a novel graph analysis task: defederation prediction, which aims to predict which edges or nodes will be removed from the graph at the next iteration, and we propose baselines for this task.
https://arxiv.org/html/2505.20882v1#S3
#feediverse #Fedivertex dataanalysis
Ric Harvey 🇪🇺🌍💚
alios
Jérémie Ducastel
RudiRastlos
EikeZiegler
ThinDragonfly
Wolfgang Lutz