Learning interpretable word embeddings via bidirectional alignment of dimensions with semantic concepts

buir.contributor.authorŞahinuç, Furkan
buir.contributor.authorÇukur, Tolga
buir.contributor.authorKoç, Aykut
buir.contributor.orcidŞahinuç, Furkan|0000-0001-9104-2860
buir.contributor.orcidÇukur, Tolga|0000-0002-2296-851X
buir.contributor.orcidKoç, Aykut|0000-0002-6348-2663
dc.citation.epage102925- 17en_US
dc.citation.issueNumber3en_US
dc.citation.spage102925- 1en_US
dc.citation.volumeNumber59en_US
dc.contributor.authorŞenel, L. K.
dc.contributor.authorŞahinuç, Furkan
dc.contributor.authorYücesoy, V.
dc.contributor.authorSchütze, H.
dc.contributor.authorÇukur, Tolga
dc.contributor.authorKoç, Aykut
dc.date.accessioned2023-02-17T10:13:45Z
dc.date.available2023-02-17T10:13:45Z
dc.date.issued2022-03-22
dc.departmentDepartment of Electrical and Electronics Engineeringen_US
dc.departmentNational Magnetic Resonance Research Center (UMRAM)en_US
dc.description.abstractWe propose bidirectional imparting or BiImp, a generalized method for aligning embedding dimensions with concepts during the embedding learning phase. While preserving the semantic structure of the embedding space, BiImp makes dimensions interpretable, which has a critical role in deciphering the black-box behavior of word embeddings. BiImp separately utilizes both directions of a vector space dimension: each direction can be assigned to a different concept. This increases the number of concepts that can be represented in the embedding space. Our experimental results demonstrate the interpretability of BiImp embeddings without making compromises on the semantic task performance. We also use BiImp to reduce gender bias in word embeddings by encoding gender-opposite concepts (e.g., male–female) in a single embedding dimension. These results highlight the potential of BiImp in reducing biases and stereotypes present in word embeddings. Furthermore, task or domain-specific interpretable word embeddings can be obtained by adjusting the corresponding word groups in embedding dimensions according to task or domain. As a result, BiImp offers wide liberty in studying word embeddings without any further effort.en_US
dc.embargo.release2024-03-22
dc.identifier.doi10.1016/j.ipm.2022.102925en_US
dc.identifier.eissn1873-5371
dc.identifier.issn0306-4573
dc.identifier.urihttp://hdl.handle.net/11693/111493
dc.language.isoEnglishen_US
dc.publisherElsevier Ltden_US
dc.relation.isversionofhttps://doi.org/10.1016/j.ipm.2022.102925en_US
dc.source.titleInformation Processing & Managementen_US
dc.subjectWord embeddingsen_US
dc.subjectInterpretabilityen_US
dc.subjectWord semanticsen_US
dc.titleLearning interpretable word embeddings via bidirectional alignment of dimensions with semantic conceptsen_US
dc.typeArticleen_US
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
Learning_interpretable_word_embeddings_via_bidirectional_alignment_of_dimensions_with_semantic_concepts.pdf
Size:
901.77 KB
Format:
Adobe Portable Document Format
Description:
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.69 KB
Format:
Item-specific license agreed upon to submission
Description: