Unsupervised detection of contextualized embedding bias with application to ideology

We propose a fully unsupervised method to detect bias in contextualized embeddings. The method leverages the assortative information latently encoded by social networks and combines orthogonality regularization, structured sparsity learning, and graph neural networks to find the embedding subspace c...

Cijeli opis

Bibliografski detalji
Glavni autori: Hofmann, V, Pierrehumbert, J, Schütze, H
Format: Conference item
Jezik:English
Izdano: Journal of Machine Learning Research 2022
_version_ 1826309038032289792
author Hofmann, V
Pierrehumbert, J
Schütze, H
author_facet Hofmann, V
Pierrehumbert, J
Schütze, H
author_sort Hofmann, V
collection OXFORD
description We propose a fully unsupervised method to detect bias in contextualized embeddings. The method leverages the assortative information latently encoded by social networks and combines orthogonality regularization, structured sparsity learning, and graph neural networks to find the embedding subspace capturing this information. As a concrete example, we focus on the phenomenon of ideological bias: we introduce the concept of an ideological subspace, show how it can be found by applying our method to online discussion forums, and present techniques to probe it. Our experiments suggest that the ideological subspace encodes abstract evaluative semantics and reflects changes in the political left-right spectrum during the presidency of Donald Trump.
first_indexed 2024-03-07T07:29:45Z
format Conference item
id oxford-uuid:ff2aa4fa-fbe2-4b2d-9d76-1a7fbec6f20e
institution University of Oxford
language English
last_indexed 2024-03-07T07:29:45Z
publishDate 2022
publisher Journal of Machine Learning Research
record_format dspace
spelling oxford-uuid:ff2aa4fa-fbe2-4b2d-9d76-1a7fbec6f20e2022-12-12T11:44:59ZUnsupervised detection of contextualized embedding bias with application to ideologyConference itemhttp://purl.org/coar/resource_type/c_5794uuid:ff2aa4fa-fbe2-4b2d-9d76-1a7fbec6f20eEnglishSymplectic ElementsJournal of Machine Learning Research2022Hofmann, VPierrehumbert, JSchütze, HWe propose a fully unsupervised method to detect bias in contextualized embeddings. The method leverages the assortative information latently encoded by social networks and combines orthogonality regularization, structured sparsity learning, and graph neural networks to find the embedding subspace capturing this information. As a concrete example, we focus on the phenomenon of ideological bias: we introduce the concept of an ideological subspace, show how it can be found by applying our method to online discussion forums, and present techniques to probe it. Our experiments suggest that the ideological subspace encodes abstract evaluative semantics and reflects changes in the political left-right spectrum during the presidency of Donald Trump.
spellingShingle Hofmann, V
Pierrehumbert, J
Schütze, H
Unsupervised detection of contextualized embedding bias with application to ideology
title Unsupervised detection of contextualized embedding bias with application to ideology
title_full Unsupervised detection of contextualized embedding bias with application to ideology
title_fullStr Unsupervised detection of contextualized embedding bias with application to ideology
title_full_unstemmed Unsupervised detection of contextualized embedding bias with application to ideology
title_short Unsupervised detection of contextualized embedding bias with application to ideology
title_sort unsupervised detection of contextualized embedding bias with application to ideology
work_keys_str_mv AT hofmannv unsuperviseddetectionofcontextualizedembeddingbiaswithapplicationtoideology
AT pierrehumbertj unsuperviseddetectionofcontextualizedembeddingbiaswithapplicationtoideology
AT schutzeh unsuperviseddetectionofcontextualizedembeddingbiaswithapplicationtoideology