mlsanityData to 7 Oct 2026

Don't Blame the Data, Blame the Model: Understanding Noise and Bias When Learning from Subjective Annotations

arXiv 2403.04085 ↗First posted 6 Mar 2024cs.CL cs.CYcited by 10

Not on the map. Why?

Abstract (arXiv)

Researchers have raised awareness about the harms of aggregating labels especially in subjective tasks that naturally contain disagreements among human annotators. In this work we show that models that are only provided aggregated labels show low confidence on high-disagreement data instances. While previous studies consider such instances as mislabeled, we argue that the reason the high-disagreement text instances have been hard-to-learn is that the conventional aggregated models underperform in extracting useful signals from subjective tasks. Inspired by recent studies demonstrating the effectiveness of learning from raw annotations, we investigate classifying using Multiple Ground Truth (Multi-GT) approaches. Our experiments show an improvement of confidence for the high-disagreement instances.

Key facts. Cited by 10 papers in the map (data through 7 Oct 2026). Builds on 33 of its 48 references that are in the map. First posted on arXiv 6 Mar 2024.

Builds on

33 of 48 references.

References · 33

  1. 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,888
  2. 2017Understanding Black-box Predictions via Influence Functionscited by 1,714
  3. 2019Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shiftcited by 1,274
  4. 2023Whose Opinions Do Language Models Reflect?cited by 759
  5. 2018An Empirical Study of Example Forgetting during Deep Neural Network Learningcited by 669
  6. 2019Data Shapley: Equitable Valuation of Data for Machine Learningcited by 617
  7. 2019A Simple Baseline for Bayesian Uncertainty in Deep Learningcited by 555
  8. 2019Deep Ensembles: A Loss Landscape Perspectivecited by 438
  9. 2023Towards Measuring the Representation of Subjective Global Opinions in Language Modelscited by 424
  10. 2020Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamicscited by 423
  11. 2019Social Bias Frames: Reasoning about Social and Power Implications of Languagecited by 377
  12. 2021Dealing with Disagreements: Looking Beyond the Majority Vote in Subjective Annotationscited by 306
21 more
  1. 2020Identifying Mislabeled Data using the Area Under the Margin Rankingcited by 259
  2. 2020Data and its (dis)contents: A survey of dataset development and use in machine learning researchcited by 259
  3. 2021Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detectioncited by 258
  4. 2019Human uncertainty makes classification more robustcited by 231
  5. 2019Understanding and Utilizing Deep Neural Networks Trained with Noisy Labelscited by 209
  6. 2019Evaluating Scalable Bayesian Deep Learning Methods for Robust Computer Visioncited by 182
  7. 2020Adversarial Filters of Dataset Biasescited by 159
  8. 2017Active Bias: Training More Accurate Neural Networks by Emphasizing High Variance Samplescited by 147
  9. 2020Continual Deep Learning by Functional Regularisation of Memorable Pastcited by 104
  10. 2018A Walk with SGDcited by 74
  11. 2021Agreeing to Disagree: Annotating Offensive Language Datasets with Annotators' Disagreementcited by 71
  12. 2020Measuring a hate speech spectrum with faceted Rasch item response theory and perspective-aware, explainable-by-design deep learningcited by 67
  13. 2022Handling Bias in Toxic Speech Detection: A Surveycited by 60
  14. 2023Everyone's Voice Matters: Quantifying Annotation Disagreement Using Demographic Informationcited by 52
  15. 2023Capturing Perspectives of Crowdsourced Annotators in Subjective Learning Taskscited by 28
  16. 2017Checkpoint Ensembles: Ensemble Methods from a Single Training Processcited by 24
  17. 2020DQI: Measuring Data Quality in NLPcited by 23
  18. 2018Active Learning for Interactive Neural Machine Translation of Data Streamscited by 20
  19. 2024Whose Emotions and Moral Sentiments Do Language Models Reflect?cited by 20
  20. 2023ACTOR: Active Learning with Annotator-specific Classification Heads to Embrace Human Label Variationcited by 17
  21. 2022Noise Audits Improve Moral Foundation Classificationcited by 7

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 4; 2025: 5; 2026: 1.

10 papers.

  1. 2025Beyond Demographics: Fine-tuning Large Language Models to Predict Individuals' Subjective Text Perceptionscited by 31
  2. 2024"Seeing the Big through the Small": Can LLMs Approximate Human Judgment Distributions on NLI from a Few Explanations?cited by 16
  3. 2024Label Convergence: Defining an Upper Performance Bound in Object Recognition through Contradictory Annotationscited by 5
  4. 2024A Collaborative Content Moderation Framework for Toxicity Detection based on Conformalized Estimates of Annotation Disagreementcited by 3
  5. 2025Looking Beyond the Obvious: A Survey on Abstract Concept Recognition for Video Understandingcited by 3
  6. 2025Conflicts in Texts: Data, Implications and Challengescited by 2
  7. 2025Are you sure? Measuring models bias in content moderation through uncertaintycited by 2
  8. 2024Behind Closed Words: Creating and Investigating the forePLay Annotated Dataset for Polish Erotic Discoursecited by 1
  9. 2025Are Humans as Brittle as Large Language Models?cited by 0
  10. 2026Learning Who Disagrees: Demographic Importance Weighting for Modeling Annotator Distributions with DiADEMcited by 0