Algorithmic Bias in BERT for Response Accuracy Prediction: A Case Study for Investigating Population Validity
Bibliographic record
Abstract
Abstract Pretrained large language models (LLMs) have gained popularity in recent years due to their high performance in various educational tasks such as learner modeling, automated scoring, automatic item generation, and prediction. Nevertheless, LLMs are black box approaches where models are less interpretable, and they may carry human biases and prejudices because historical human data have been used for pretraining these large‐scale models. For these reasons, the prediction tasks based on LLMs require scrutiny to ensure that the prediction models are fair and unbiased. In this study, we used BERT—a pretrained encoder‐only LLM for predicting response accuracy using action sequences extracted from the 2012 PIAAC assessment. We selected three countries (i.e., Finland, Slovakia, and the United States) representing different performance levels in the overall PIAAC assessment. We found promising results for predicting response accuracy using the fine‐tuned BERT model. Additionally, we examined algorithmic bias in the prediction models trained with different countries. We found differences in model performance, suggesting that some trained models are not free from bias, and thus the models are less generalizable across countries. Our results highlighted the importance of investigating algorithmic fairness in prediction models utilizing algorithmic systems to ensure models are bias‐free.
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 | 0.053 | 0.185 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
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".