Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems
Open accessAbstract
Automatic machine learning systems can inadvertently accentuate and perpetuate inappropriate human biases.Past work on examining inappropriate biases has largely focused on just individual systems.Further, there is no benchmark dataset for examining inappropriate biases in systems.Here for the first time, we present the Equity Evaluation Corpus (EEC), which consists of 8,640 English sentences carefully chosen to tease out biases towards certain races and genders.We use the dataset to examine 219 automatic sentiment analysis systems that took part in a recent shared task, SemEval-2018 Task 1 'Affect in Tweets'.We find that several of the systems show statistically significant bias; that is, they consistently provide slightly higher sentiment intensity predictions for one race or one gender.We make the EEC freely available.
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
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?
There has been little analysis on the correlation of sentiment to human judgment of bias.
“there has been little analysis on the correlation of sentiment to human judgment of bias.”
From The Woman Worked as a Babysitter: On Biases in Language Generation (2019) · cites this paper as Kiritchenko and Mohammad, 2018
The Related Work section says that bias in sentiment analysis systems had received limited study. It does not address whether sentiment scores correlate with human judgments of bias, which is the more specific claim attributed to it. (LLM, conf=0.82)
Experimental — accuracy not yet measuredEquity Evaluation Corpus sentences are generated from templates instantiated with person and emotional state words, with tuples containing the same words except for the person term.
“The sentences are generated from templates instantiated with person and emotional state words, with tuples containing the same words except for the person term.”
From Bias and Fairness in Large Language Models: A Survey (2024) · cites this paper as Kiritchenko and Mohammad 2018
Section 3 (The Equity Evaluation Corpus) says the sentences were generated from templates using person terms and that pairs can differ only in the gender- or race-associated person term. But emotional-state words occur only in templates 1–4: templates 5–7 use words describing situations or events, and templates 8–11 have no emotion words. The claim therefore overgeneralizes the emotional-state-word component. (LLM, conf=0.88)
Experimental — accuracy not yet measuredEquity Evaluation Corpus contains 8,640 sentences to measure differences in sentiment towards gender and racial groups.
“Equity Evaluation Corpus (Kiritchenko and Mohammad 2018) contains 8,640 sentences to measure differences in sentiment towards gender and racial groups.”
From Bias and Fairness in Large Language Models: A Survey (2024) · cites this paper as Kiritchenko and Mohammad 2018
Performances of sentiment analysis of male- and female-authored texts are extremely variable and potentially biased.
“performances of sentiment analysis of male- and female-authored texts are extremely variable80 and potentially biased81.”
From Sex and gender differences and biases in artificial intelligence for biomedicine and healthcare (2020) · cites this paper as [81]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 35 · Cited by 384
Research works printed but not linked: 2
- Amit Datta, Michael Carl Tschantz, and Anupam Datta. 2015. Automated experiments on ad privacy settings. Proceedings on Privacy Enhancing Tech- nologies, 2015(1):92-112.
- Saif M. Mohammad. 2018. Word affect intensi- ties. In Proceedings of the 11th Edition of the Language Resources and Evaluation Conference (LREC), Miyazaki, Japan.
- Authors
- Kiritchenko, S., & Mohammad, S.
- Affiliations
- National Research Council Canada (CA)
- Venue
- Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics
- Published
- 2018
- Pages
- 43-53
- OpenAlex
- W2802105481
- Cited by
- 384
- References
- 35
- Citations by year
- 2026: 25 · 2025: 36 · 2024: 40 · 2023: 64 · 2022: 64 · 2021: 72 · 2020: 45 · 2019: 34 · 2018: 3 · 2013: 1
- FWCI
- 68.88
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Joint Conference on Lexical and Computational Semantics (publishedVersion, cc-by)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Keywords
- sentiment analysis, gender bias, racial bias, algorithmic bias, 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
