Small or Large? Zero-Shot or Finetuned? Guiding Language Model Choice for Specialized Applications in Healthcare
Bibliographic record
Abstract
Objectives: To guide language model (LM) selection by comparing finetuning vs. zero-shot use, generic pretraining vs. domain-adjacent vs. further domain-specific pretraining, and bidirectional language models (BiLMs) such as BERT vs. unidirectional LMs (LLMs) for clinical classification. Materials and Methods: We evaluated BiLMs (RoBERTa, PathologyBERT, Gatortron) and LLM (Mistral nemo instruct 12B) on three British Columbia Cancer Registry (BCCR) pathology classification tasks varying in difficulty/data size. We assessed zero-shot vs. finetuned BiLMs, zero-shot LLM, and further BCCR-specific pretraining using macro-average F1 scores. Results: Finetuned BiLMs outperformed zero-shot BiLMs and zero-shot LLM. The zero-shot LLM outperformed zero-shot BiLMs but was consistently outperformed by finetuned BiLMs. Domain-adjacent BiLMs generally outperformed generic BiLMs after finetuning. Further domain-specific pretraining boosted complex/low-data task performance, with otherwise modest gains. Conclusions: For specialized classification, finetuning BiLMs is crucial, often surpassing zero-shot LLMs. Domain-adjacent pretrained models are recommended. Further domain-specific pretraining provides significant performance boosts, especially for complex/low-data scenarios. BiLMs remain relevant, offering strong performance/resource balance for targeted clinical tasks.
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 | 0.004 | 0.017 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
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".