HateCheck: functional tests for hate speech detection models
Detecting online hate is a difficult task that even state-of-the-art models struggle with. Typically, hate speech detection models are evaluated by measuring their performance on held-out test data using metrics such as accuracy and F1 score. However, this approach makes it difficult to identify spe...
| Main Authors: | , , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Association for Computational Linguistics
2021
|
| _version_ | 1826311950786625536 |
|---|---|
| author | Röttger, P Vidgen, B Dong, N Waseem, Z Margetts, H Pierrehumbert, JB |
| author_facet | Röttger, P Vidgen, B Dong, N Waseem, Z Margetts, H Pierrehumbert, JB |
| author_sort | Röttger, P |
| collection | OXFORD |
| description | Detecting online hate is a difficult task that even state-of-the-art models struggle with. Typically, hate speech detection models are evaluated by measuring their performance on held-out test data using metrics such as accuracy and F1 score. However, this approach makes it difficult to identify specific model weak points. It also risks overestimating generalisable model performance due to increasingly well-evidenced systematic gaps and biases in hate speech datasets. To enable more targeted diagnostic insights, we introduce HateCheck, a suite of functional tests for hate speech detection models. We specify 29 model functionalities motivated by a review of previous research and a series of interviews with civil society stakeholders. We craft test cases for each functionality and validate their quality through a structured annotation process. To illustrate HateCheck’s utility, we test near-state-of-the-art transformer models as well as two popular commercial models, revealing critical model weaknesses. |
| first_indexed | 2024-03-07T08:18:51Z |
| format | Conference item |
| id | oxford-uuid:87f02544-2770-4d4a-a143-c1be682e7f95 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:18:51Z |
| publishDate | 2021 |
| publisher | Association for Computational Linguistics |
| record_format | dspace |
| spelling | oxford-uuid:87f02544-2770-4d4a-a143-c1be682e7f952024-01-16T14:24:54ZHateCheck: functional tests for hate speech detection modelsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:87f02544-2770-4d4a-a143-c1be682e7f95EnglishSymplectic ElementsAssociation for Computational Linguistics2021Röttger, PVidgen, BDong, NWaseem, ZMargetts, HPierrehumbert, JBDetecting online hate is a difficult task that even state-of-the-art models struggle with. Typically, hate speech detection models are evaluated by measuring their performance on held-out test data using metrics such as accuracy and F1 score. However, this approach makes it difficult to identify specific model weak points. It also risks overestimating generalisable model performance due to increasingly well-evidenced systematic gaps and biases in hate speech datasets. To enable more targeted diagnostic insights, we introduce HateCheck, a suite of functional tests for hate speech detection models. We specify 29 model functionalities motivated by a review of previous research and a series of interviews with civil society stakeholders. We craft test cases for each functionality and validate their quality through a structured annotation process. To illustrate HateCheck’s utility, we test near-state-of-the-art transformer models as well as two popular commercial models, revealing critical model weaknesses. |
| spellingShingle | Röttger, P Vidgen, B Dong, N Waseem, Z Margetts, H Pierrehumbert, JB HateCheck: functional tests for hate speech detection models |
| title | HateCheck: functional tests for hate speech detection models |
| title_full | HateCheck: functional tests for hate speech detection models |
| title_fullStr | HateCheck: functional tests for hate speech detection models |
| title_full_unstemmed | HateCheck: functional tests for hate speech detection models |
| title_short | HateCheck: functional tests for hate speech detection models |
| title_sort | hatecheck functional tests for hate speech detection models |
| work_keys_str_mv | AT rottgerp hatecheckfunctionaltestsforhatespeechdetectionmodels AT vidgenb hatecheckfunctionaltestsforhatespeechdetectionmodels AT dongn hatecheckfunctionaltestsforhatespeechdetectionmodels AT waseemz hatecheckfunctionaltestsforhatespeechdetectionmodels AT margettsh hatecheckfunctionaltestsforhatespeechdetectionmodels AT pierrehumbertjb hatecheckfunctionaltestsforhatespeechdetectionmodels |