Bhavyagowni/bart

Model Card for Model ID

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Developed by: Gowni Bhavishya,Dr.Shib Shankar sahu

• Model type: Sequence-to-Sequence model (BART) fine-tuned for scientific highlight generation

• Language(s) (NLP): English

• Finetuned from model [optional]: facebook/bart-large.

• Repository: [More Information Needed]

Uses

1. Researchers in biomedical and scientific fields
2. Academic publishers and editors
3. Developers building scientific summarization tools
4. NLP practitioners working on domain-specific summarization

Direct Use

Generate highlights or concise summaries of scientific abstracts (especially biomedical, life sciences, or clinical research)

[More Information Needed]

Out-of-Scope Use

1. Not suitable for general news summarization, social media content, or informal language.
2. Should not be used for critical medical decision-making or clinical diagnostics.
3. Not designed for creative writing, dialogue generation, or question answering.
4. Avoid using this model for non-English abstracts or multilingual input—it was trained on English biomedical text only.

Bias, Risks, and Limitations

While BART performs well on biomedical abstracts, it inherits limitations from both:

1. Pretrained BART model biases (from general corpora like Wikipedia and Books)
2. Training dataset distribution biases (e.g., if your abstracts are from PubMed or a niche field) Known Limitations:
3. May generate generic summaries if abstracts are vague or long.
4. Struggles with mathematical, chemical, or symbolic notation.
5. Output may appear plausible but factually incorrect.
6. Does not provide citations or references for claims.

Recommendations

1. Always validate generated summaries against the full abstract or ground truth highlights.
2. Preferably use in human-in-the-loop systems where an expert reviews the output.
3. Fine-tune further or filter input for domain-specific tasks (e.g., cardiology vs oncology). Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

Training Details

Training Data

1.Fine-tuned on a dataset of scientific abstracts and their corresponding highlights. The training dataset was split into train (10k), validation (2k), and test (1.8k) sets. Input: Abstract column Target: Highlights column (only in train/val)

Training Hyperparameters

Model architecture: facebook/bart-large Batch size: 4 (per device) Epochs: 5 Learning rate: 2e-5

Evaluation

Rouge1,Rouge2,RougeL,Meteor.

Testing Data, Factors & Metrics

Testing Data

The test set consists of 1,840 scientific abstracts without ground-truth highlights.

Metrics

ROUGE-1: Measures unigram overlap (precision & recall) ROUGE-2: Measures bigram overlap ROUGE-L: Measures longest common subsequence METEOR: Incorporates synonymy, stemming, and word order

Results

Summary

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

More Information [optional]

SVNIT CSE

Model Card Authors [optional]

Gowni Bhavishya,Dr.Shib Sankar Sahu

Model Card Contact

[More Information Needed]