| Challenge: | Existing work investigating social bias in factual knowledge graphs has focused on knowledge graph embeddings, so more recent classes of models achieving superior results by fine-tuning Transformers have not yet been investigated. |
| Approach: | They propose a model-agnostic approach for bias measurement leveraging fairness metrics to compare bias in knowledge graph embedding-based predictions (KG only) with models that use pre-trained, Transformer-based language models (KG+LM). |
| Outcome: | The proposed model-agnostic approach compares gender bias in occupation predictions with models that use pre-trained, Transformer-based language models (KG+LM). |
Similar Papers
Towards Automatic Bias Detection in Knowledge Graphs (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have shown that knowledge graphs are prone to various social biases, and have proposed multiple methods for debiasing them. |
| Approach: | They propose a framework for identifying biases present in knowledge graph embeddings based on numerical bias metrics. |
| Outcome: | The proposed framework can be extended to further bias definitions and applications. |
Measuring Fairness with Biased Rulers: A Comparative Study on Bias Metrics for Pre-trained Language Models (2022.naacl-main)
Copied to clipboard
| Challenge: | An increasing awareness of biased patterns in natural language processing resources such as BERT has motivated many metrics to quantify ‘bias’ and ‘fairness’. |
| Approach: | They combine literature survey, correlation analysis and empirical evaluations to evaluate compatibility of fairness metrics for pre-trained language models and their downstream tasks. |
| Outcome: | The proposed measures are not compatible with each other and highly depend on (i) templates, (ii) attribute and target seeds and (iv) the choice of embeddings. |
Debiasing knowledge graph embeddings (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to train knowledge graph embeddings to be neutral to sensitive attributes such as gender have been shown to increase training time by a factor of eight or more. |
| Approach: | They propose a method where all embeddings are trained to be neutral to sensitive attributes such as gender by default using an adversarial loss. |
| Outcome: | The proposed method reduces training time by eightfold and improves accuracy. |
Predictive Multiplicity of Knowledge Graph Embeddings in Link Prediction (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Knowledge graph embeddings (KGE) models are often used to predict missing links for knowledge graphs (KGs) however, multiple KG embedds can give conflicting predictions for unseen queries. |
| Approach: | They define predictive multiplicity in link prediction and introduce evaluation metrics to measure it using commonly used benchmark datasets. |
| Outcome: | The proposed methods significantly mitigat conflicts by 66% to 78% in link prediction. |
Measuring Social Biases in Masked Language Models by Proxy of Prediction Quality (2025.acl-long)
Copied to clipboard
| Challenge: | Innovative transformer-based language models produce contextually-aware token embeddings, but have been shown to encode unwanted biases for downstream applications. |
| Approach: | They extend previous work by evaluating social biases introduced after retraining an MLM under the masked language modeling objective and propose proxy functions within an iterative masking experiment to measure the quality of transformer models’ predictions. |
| Outcome: | The proposed proxy functions within an iterative masking experiment show that all transformer models encode concerning social biases. |
Quantifying Social Biases in NLP: A Generalization and Empirical Comparison of Extrinsic Fairness Metrics (2021.tacl-1)
Copied to clipboard
| Challenge: | Existing fairness metrics quantify the differences in a model’s behaviour across a range of demographic groups. |
| Approach: | They propose to unify existing fairness metrics and compare them to three generalized fairness measures to reveal the connections between them. |
| Outcome: | The proposed measures can be explained by differences in parameter choices, and the results are consistent with previous studies. |
Bias and Fairness in Natural Language Processing (D19-2)
Copied to clipboard
| Challenge: | a tutorial will review the history of bias and fairness studies in machine learning and language processing . |
| Approach: | This tutorial reviews the history of bias and fairness studies in machine learning and language processing . it presents recent community effort to quantify and mitigat bias in natural language processing models . |
| Outcome: | This tutorial reviews the history of bias and fairness studies in machine learning and language processing . it aims to quantify and mitigate bias in natural language processing models for a wide spectrum of tasks . |
GKnow: Measuring the Entanglement of Gender Bias and Factual Gender (2026.acl-long)
Copied to clipboard
| Challenge: | Recent studies have focused on mitigating gender bias, but mechanistic interpretations of gender fail to distinguish between factually gendered outputs and gender biased outputs. |
| Approach: | They propose a benchmark to assess gender knowledge and gender bias in language models . they use neuron ablation to disentangle stereotypical and factual gender . |
| Outcome: | The proposed benchmark assesses gender knowledge and gender bias in language models across different types of gender-related predictions. |
Are you sure? Measuring models bias in content moderation through uncertainty (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Language Model-based classifiers perpetuate racial and social biases in content moderation . et al., j. n. d., and j neil, e. c. (2005) measure the fairness of content moderated models . |
| Approach: | They propose an unsupervised approach that benchmarks models on their uncertainty . they use uncertainty as a proxy to analyze the bias of 11 models against women and non-whites . |
| Outcome: | The proposed method analyzes the bias of 11 models against women and non-white annotators . it shows that some pre-trained models predict with high accuracy the labels coming from minority groups . |
Benchmarking Intersectional Biases in NLP (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent work on fairness of machine learning models has focused on how to debias, but research on the fairness and performance of biased/debiased models on downstream prediction tasks has been limited. |
| Approach: | They assess intersectional bias - fairness across multiple demographic dimensions . they highlight possible causes and make recommendations for future NLP debiasing research. |
| Outcome: | The proposed approaches fare well in terms of fairness-accuracy trade-off, but are unable to effectively alleviate bias in downstream tasks. |