README.md

---
language: en
license: apache-2.0
base_model: allenai/scibert_scivocab_uncased
pipeline_tag: text-classification
library_name: transformers
tags:
  - scientific-text
  - citation-intent
datasets:
  - allenai/multicite
  - allenai/scicite
widget:
  - text: "Deep learning has achieved remarkable success in many NLP tasks [1, 2]."
    example_title: "Background citation"
  - text: "We adopt the evaluation protocol proposed by Smith et al. (2020)."
    example_title: "Non-background citation"
---

# SciBERT Citation-Background Classifier

[`allenai/scibert_scivocab_uncased`](https://huggingface.co/allenai/scibert_scivocab_uncased) fine-tuned as a binary classifier of citation intent: does a citation sentence cite prior work as background (`BACKGROUND`) or for any other reason (`NOT_BACKGROUND`)?

## Labels

| id | label | meaning |
|----|----------------|--------------------------------------------------------------|
| 0 | NOT_BACKGROUND | citation used for method, comparison, extension, motivation… |
| 1 | BACKGROUND | citation provides background/context for the citing paper |

## How to use

```python
from transformers import pipeline

clf = pipeline("text-classification", model="hongccccccc/scibert-citation-background-classifier")
clf("Deep learning has achieved remarkable success in many NLP tasks [1, 2].")
# [{'label': 'BACKGROUND', 'score': ...}]
```

## Training

- **Base model:** SciBERT (uncased, scivocab; `BertForSequenceClassification`, single-label, 2 classes)
- **Data:** the combined annotated citation contexts of [SciCite](https://github.com/allenai/scicite) (Cohan et al., 2019) and [MultiCite](https://github.com/allenai/multicite) (Lauscher et al., 2022) — 27,052 instances, 11,635 (43%) labeled `BACKGROUND`
- **Fine-tuned:** August 2023, `transformers` 4.32.0

## Evaluation

F1 score of **0.81** on a held-out test set (80-20 train-test split of the combined dataset).