nlpub / chinese-whispers

An implementation of Chinese Whispers in Python.
https://chinese-whispers.readthedocs.io
MIT License
58 stars 13 forks source link
chinese-whispers clustering graph graph-clustering networkx python

Chinese Whispers for Python

This is an implementation of the Chinese Whispers clustering algorithm in Python. Since this library is based on NetworkX, it is simple to use.

Unit Tests Read the Docs PyPI Version Conda Version

Installation

Usage

Given a NetworkX graph G, this library can cluster it using the following code:

from chinese_whispers import chinese_whispers
chinese_whispers(G, weighting='top', iterations=20)

As the result, each node of the input graph is provided with the label attribute that stores the cluster label.

The library also offers a convenient command-line interface (CLI) for clustering graphs represented in the ABC tab-separated format (source\ttarget\tweight).

# Write karate_club.tsv (just as example)
python3 -c 'import networkx as nx; nx.write_weighted_edgelist(nx.karate_club_graph(), "karate_club.tsv", delimiter="\t")'

# Using as CLI
chinese-whispers karate_club.tsv

# Using as module (same CLI as above)
python3 -mchinese_whispers karate_club.tsv

A more complete usage example is available in the example notebook and at https://nlpub.github.io/chinese-whispers/.

In case you require higher performance, please consider our Java implementation that also includes other graph clustering algorithms: https://github.com/nlpub/watset-java.

Citation

@article{Ustalov:19:cl,
  author    = {Ustalov, Dmitry and Panchenko, Alexander and Biemann, Chris and Ponzetto, Simone Paolo},
  title     = {{Watset: Local-Global Graph Clustering with Applications in Sense and Frame Induction}},
  journal   = {Computational Linguistics},
  year      = {2019},
  volume    = {45},
  number    = {3},
  pages     = {423--479},
  doi       = {10.1162/COLI_a_00354},
  publisher = {MIT Press},
  issn      = {0891-2017},
  language  = {english},
}

Copyright

Copyright (c) 2018–2024 Dmitry Ustalov. See LICENSE for details.