Skip to content
This repository has been archived by the owner on Jul 10, 2019. It is now read-only.
/ watset Public archive

Watset: Automatic Induction of Synsets from a Graph of Synonyms

License

Notifications You must be signed in to change notification settings

dustalov/watset

Repository files navigation

Watset: Automatic Induction of Synsets from a Graph of Synonyms

Watset is a local-global meta-algorithm for fuzzy graph clustering. The algorithm constructs an intermediate representation, called a sense graph, using a local graph clustering algorithm and then obtains overlapping node clusters using a global graph clustering algorithm.

Originally, Watset was designed for addressing the synset induction problem, which is indicated in the corresponding ACL 2017 paper. Despite its simplicity, Watset shows excellent results, outperforming five competitive state-of-the-art methods in terms of F-score on four gold standard datasets for English and Russian derived from large-scale manually constructed lexical resources.

Important

We found that Watset works very well not just for synset induction, but for many other fuzzy clustering tasks, too. Please use a much faster and convenient implementation of Watset in Java: 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},
  doi       = {10.1162/COLI_a_00354},
  publisher = {MIT Press},
  issn      = {0891-2017},
  language  = {english},
}

Copyright

This repository contains the implementation of Watset. See LICENSE for details.