Bias Mitigation in Nlp Models Using Bert
Bibliographic record
Abstract
Natural Language Processing (NLP) models, despite being popularly applied to a variety of applications, tend to inherit and exacerbate biases in their training data. These biases, especially social bias, can takes place because of skewed or non-representative training data, causing discriminatory or unfair model behavior. To address this, this research proposes bias removal in language models using adversarial fine-tuning of BERT on the CrowS-Pairs dataset, which is a state-of-the-art benchmark for identifying social bias in text. The method contains training of dedicated model that pairs masked language modeling (MLM) with an auxiliary adversarial classification task. The model is trained specifically to conduct standard language modeling while decreasing its capacity to discriminate between biased and debiased sentence pairs. The specified loss function brings the two into balance, calling for the model to learn discriminatively linguistic features. The pipeline used for training incorporates a dataset handler, a tokenizer, as well as data collator which is optimized to support MLM learning tasks. Multipe epochs experiment is used while training, with the finetuned model preserved for subsequent experimentation. This two-task learning paradigm illustrates the promise of adversarial goals in minimizing unwanted biases in NLP models without diminishing their language comprehension abilities.
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 | 0.003 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 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".