Compressed models for co-reference resolution: enhancing efficiency with debiased word embeddings
Compressed models for co-reference resolution: enhancing efficiency with debiased word embeddings
About this item
Full title
Author / Creator
Publisher
London: Nature Publishing Group UK
Journal title
Language
English
Formats
Publication information
Publisher
London: Nature Publishing Group UK
Subjects
More information
Scope and Contents
Contents
This work presents a comprehensive approach to reduce bias in word embedding vectors and evaluate the impact on various Natural Language Processing (NLP) tasks. Two GloVe variations (840B and 50) are debiased by identifying the gender direction in the word embedding space and then removing or reducing the gender component from the embeddings of tar...
Alternative Titles
Full title
Compressed models for co-reference resolution: enhancing efficiency with debiased word embeddings
Authors, Artists and Contributors
Author / Creator
Identifiers
Primary Identifiers
Record Identifier
TN_cdi_doaj_primary_oai_doaj_org_article_87033d2a03fd4e73b5043d2b0f03e7db
Permalink
https://devfeature-collection.sl.nsw.gov.au/record/TN_cdi_doaj_primary_oai_doaj_org_article_87033d2a03fd4e73b5043d2b0f03e7db
Other Identifiers
ISSN
2045-2322
E-ISSN
2045-2322
DOI
10.1038/s41598-023-45677-0
