/wsd-com-sem

Primary LanguageHTML

wsd-com-sem

Code for Computational Semantics project on Word Sense Disambiguation (WSD).

Authors: M. Bouma (S3142558), F. Perin (S2865300) at MSc AI, Rijksuniversiteit Groningen

running experiments

To install the required packages, run: pip install -r requirements.txt

Then run: python online_window.py --feat ('sym'|'sem'| 'sns')+ -k K --bucket_size (3| 5| 7)+ --pooling <voting|average>

Notes:

  • do not use the --pooling parameter when --bucket_size has a single value.
  • -k is a float value; we recommend assigning it between 0 and 10.
  • adjust pip and python if needed to use Python 3.

data files

  • txt: dev, train, and test files in the form of txt, from the conll data provided.
  • csv: dev, train, and test files in the form of csv, with cat and rol columns removed.
  • context: csv files of context windows from early version, based on dev.

dev_files

Files, code, notebooks, etc. used for development that are not relevant to any other user.

data_exp files

  • data_exp.ipynb: jupyter notebook describing data exploration.
  • data_exp.html: html version of the notebook for easier viewing.

online_window.py

Contains functions related to the scrolling context window, evaluation, and the main function of the code.

count.py

Contains functions related to counting frequencies, feature extraction from words in contexts, probabilities and calculation.