RoBERTa Abstract-Section Classifier
roberta-base fine-tuned to classify sentences from scientific abstracts into five rhetorical sections: BACKGROUND, CONCLUSIONS, METHODS, OBJECTIVE, RESULTS.
Labels
| id | label | example |
|---|---|---|
| 0 | BACKGROUND | "Diabetes is a growing public health concern…" |
| 1 | CONCLUSIONS | "We conclude that early intervention improves outcomes." |
| 2 | METHODS | "Participants were randomly assigned to two groups…" |
| 3 | OBJECTIVE | "The aim of this study was to evaluate…" |
| 4 | RESULTS | "The treatment group showed a 40% reduction (p < 0.001)." |
How to use
from transformers import pipeline
clf = pipeline("text-classification", model="hongccccccc/roberta-abstract-section-classifier")
clf("The treatment group showed a 40% reduction in mortality compared with placebo.")
# [{'label': 'RESULTS', 'score': 0.99}]
Training
- Base model: RoBERTa (Liu et al., 2019;
RobertaForSequenceClassification, single-label, 5 classes) - Data: 200,000 paper abstracts from PubMed (Canese and Weis, 2013), self-labeled with the five section categories
- Fine-tuned: January 2023,
transformers4.12.5 (originaltraining_args.binincluded)
Evaluation
F1 score of 0.92 on a held-out 10% sample (details in Wright et al., 2022).