Semantic spaces module

This is a python module that allows to compute semantic metrics based on distributional semantics models.

For example, to find words that are semantically similar to the word 'brain':

from semspaces.space import SemanticSpace

space = SemanticSpace.from_csv('space.w2v.gz')

space.most_similar(['brain'])

{'brain': [(u'brain', 0.0),
  (u'brains', 0.34469844325620635),
  (u'cerebrum', 0.4426992023455152),
  (u'cerebellum', 0.4483798859566903),
  (u'cortical', 0.469348588934828),
  (u'brainstem', 0.4791188497952641),
  (u'cortex', 0.479544888313173),
  (u'ganglion', 0.49717579235842546),
  (u'thalamus', 0.5030885466349713),
  (u'thalamic', 0.5059524199702277)]}

The module wraps dense and sparse matrix implementations to provide convenience methods for computing semantic statistics as well as easy input and output of the data.

Installation

pip install -r requirements.txt
python setup.py install

Semantic spaces

You can download a set of validated semantic spaces for English and Dutch here (see Mandera, Keuleers, & Brysbaert, in press).

Contribute

• Issue Tracker: https://github.com/pmandera/semspaces/issues
• Source Code: https://github.com/pmandera/semspaces

Authors

The tool was developed at Center for Reading Research, Ghent University by Paweł Mandera.

License

The project is licensed under the Apache License 2.0.

References

Mandera, P., Keuleers, E., & Brysbaert, M. (in press). Explaining human performance in psycholinguistic tasks with models of semantic similarity based on prediction and counting: A review and empirical validation. Journal of Memory and Language.