Cross-domain fairness audit of sentiment label bias in foundation models: Comparing human and machine annotations on tweets and reviews
Bibliographic record
Abstract
This study presents a comparative fairness audit of leading foundation models (FMs), OpenAI, Gemini, DeepSeek, and LLaMA, against human labeled sentiment data across diverse text domains. Using tweet, review, and sarcasm labeled datasets, we assess model agreement, fairness metrics (e.g., Demographic Parity Difference, Equal Opportunity Difference, Disparate Impact Ratio), and statistical significance of label discrepancies. Our findings reveal performance gaps, domain sensitivity, and systematic biases, especially in sarcastic and informal texts. To address these biases, we conduct a pilot sarcasm aware multitask fine tuning experiment, which reduces misclassification disparities and improves fairness metrics on sarcastic samples. These results underscore both the necessity of fairness audits and the potential of lightweight mitigation strategies in sentiment classification tasks. • Foundation models show significant disparity in sentiment predictions across domains and labels. • Performance drops sharply on sarcastic text, revealing weaknesses in nuance detection. • Misclassification and disagreement patterns highlight the need for fairness-aware sentiment tuning.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.042 | 0.099 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".