---
title: Article Topic Service
sdk: docker
app_port: 7860
pinned: false
license: mit
---

# Article Topic Service

Streamlit demo for a SciBERT classifier that predicts the topic of a scientific article from its title and abstract.

The app returns the top class together with the smallest set of labels whose cumulative probability exceeds 95%.
It also supports PDF upload with automatic extraction of the article title and abstract from the first pages of the document.

Repository extras:

- `Example pdf article.pdf` is a sample PDF that the app can parse and classify.
- `Training/Training readme.md` contains the short training report.
- `Training/prepare_dataset.py`, `Training/train_classifier.py`, and `Training/service_checks.py` contain the dataset preparation, training, and validation code.