The DOI matches its Crossref record; 12 of the 12 references it cites were found in a registry or an index.
Measuring and Mitigating Unintended Bias in Text Classification
Open accessAbstract
We introduce and illustrate a new approach to measuring and mitigating unintended bias in machine learning models. Our definition of unintended bias is parameterized by a test set and a subset of input features. We illustrate how this can be used to evaluate text classifiers using a synthetic test set and a public corpus of comments annotated for toxicity from Wikipedia Talk pages. We also demonstrate how imbalances in training data can lead to unintended bias in the resulting models, and therefore potentially unfair applications. We use a set of common demographic identity terms as the subset of input features on which we measure bias. This technique permits analysis in the common scenario where demographic information on authors and readers is unavailable, so that bias mitigation must focus on the content of the text itself. The mitigation method we introduce is an unsupervised approach based on balancing the training dataset. We demonstrate that this approach reduces the unintended bias without compromising overall model quality.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Synthetic data is valuable here because [11] shows that real data often has disproportionate amounts of toxicity directed at specific groups.”
From Model Cards for Model Reporting (2019) · cites this paper as [11]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Prior work has shown that toxicity classifiers can pick up on biases in the training data and spuriously associate toxicity with the mention of certain demographics (Park et al., 2018; Dixon et al., 2018).”
From WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020) · cites this paper as Dixon et al., 2018
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
An introduction to ROC analysis
Fawcett, T. (2006) · Pattern Recognition Letters · doi:10.1016/j.patrec.2005.10.010 · Cited by 21,879
Equality of Opportunity in Supervised Learning
Hardt, M., Price, E., & Srebro, N. (2016) · arXiv · doi:10.48550/arxiv.1610.02413 · Cited by 2,536
Certifying and Removing Disparate Impact
Feldman, M., Friedler, S. A., Moeller, J., Scheidegger, C., & Venkatasubramanian, S. (2015) · Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2783258.2783311 · Cited by 1,827
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016) · arXiv · doi:10.48550/arxiv.1607.06520 · Cited by 1,615
Inherent Trade-Offs in the Fair Determination of Risk Scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016) · arXiv · doi:10.48550/arxiv.1609.05807 · Cited by 941
Ex Machina: Personal Attacks Seen at Scale
Wulczyn, E., Thain, N., & Dixon, L. (2017) · Proceedings of the 26th International Conference on World Wide Web · doi:10.1145/3038912.3052591 · Cited by 636
Self-citationInherent Tradeoffs in the Fair Determination of Risk Scores
Raghavan, M. (2023) · ACM eBooks · doi:10.1145/3603195.3603197 · Cited by 350
Gender and Dialect Bias in YouTube’s Automatic Captions
Tatman, R. (2017) · Proceedings of the First ACL Workshop on Ethics in Natural Language Processing · doi:10.18653/v1/w17-1606 · Cited by 338
The social impact of natural language processing
Hovy, D., & Spruit, S. L. (2016) · Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) · doi:10.18653/v1/p16-2096 · Cited by 329
Data Decisions and Theoretical Implications when Adversarially Learning Fair RepresentationsPreprint
Beutel, A., Chen, J., Zhao, Z., & Chi, E. H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.00075 · Cited by 282
On the (im)possibility of fairness
Friedler, S. A., Scheidegger, C., & Venkatasubramanian, S. (2016) · arXiv · doi:10.48550/arxiv.1609.07236 · Cited by 247
Blodgett, S. L., & O'Connor, B. (2017) · arXiv · doi:10.48550/arxiv.1707.00061 · Cited by 153
Cited by
Sorted by citations
Model Cards for Model Reporting
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287596 · Cited by 2,681
“A model card could even extend beyond these summary statistics to reveal other measures of differences between distributions such as cross entropy, perplexity, KL divergence and pinned area under the curve (pinned AUC) [13].”
BackgroundMethodInfluentialSelf-citationPaLM: Scaling Language Modeling with Pathways
Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.02311 · Cited by 2,590
“We note that we are subject to the social bias of the Perspective API, which can assign higher toxicity to innocuous mentions of certain identity groups (Dixon et al., 2018; Röttger et al., 2021).”
BackgroundInfluentialLanguage (Technology) is Power: A Critical Survey of “Bias” in NLP
Blodgett, S. L., Barocas, S., Daumé III, H., & Wallach, H. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.485 · Cited by 1,050
“A.2 Full categorization: Motivations Allocational harms Hovy and Spruit (2016); Caliskan et al. (2017); Madnani et al. (2017); Dixon et al. (2018); Kiritchenko and Mohammad (2018); Shen et al. (2018); Zhao et al. (2018b); Bhaskaran and Bhallamudi (2019); Bordia and Bowman (2019); Brunet et al.…”
BackgroundInfluentialThe Risk of Racial Bias in Hate Speech Detection
Sap, M., Card, D., Gabriel, S., Choi, Y., & Smith, N. A. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1163 · Cited by 817
Taxonomy of Risks posed by Language Models
Weidinger, L., Uesato, J., Rauh, M., Griffin, C., Huang, P.-S., Mellor, J., Glaese, A., Cheng, M., Balle, B., Kasirzadeh, A., et al. (2022) · 2022 ACM Conference on Fairness Accountability and Transparency · doi:10.1145/3531146.3533088 · Cited by 764
Metadata
- Authors
- Dixon, L., Li, J., Sorensen, J., Thain, N., & Vasserman, L.
- Affiliations
- Google (United States) (US)
- Venue
- Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society
- Publisher
- ACM
- Published
- 2018-12-27
- Pages
- 67-73
- ISSN
- 3065-8365 (ISSN-L)
- OpenAlex
- W2791170418
- Cited by
- 674
- References
- 11
- Citations by year
- 2026: 43 · 2025: 77 · 2024: 70 · 2023: 103 · 2022: 120 · 2021: 132 · 2020: 76 · 2019: 50 · 2018: 3
- FWCI
- 115.48
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- Proceedings of the AAAI/ACM Conference on AI Ethics and Society (publishedVersion, cc-by-nc-nd)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Imbalanced Data Classification Techniques
- Keywords
- text classification, bias mitigation, demographic bias, toxicity detection, Wikipedia talk pages, imbalanced training data, algorithmic fairness
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
