README.md

---
title: ArXiv CS Sub-field Classifier
emoji: 📚
colorFrom: indigo
colorTo: pink
sdk: gradio
sdk_version: 4.44.1
python_version: "3.11"
app_file: app.py
pinned: false
license: apache-2.0
---

# arXiv CS Sub-field Classifier — Demo Space

Paste an arXiv abstract; the model returns a top-k prediction over six
overlapping CS sub-fields (`cl_nlp`, `lg_ml`, `cv_vision`, `ir_retrieval`,
`ai_general`, `stat_ml`).

- **Model**: [Tiansve/arxiv-subfield-linear-probe](https://huggingface.co/Tiansve/arxiv-subfield-linear-probe)
- **Dataset**: [Tiansve/arxiv-cs-subfield](https://huggingface.co/datasets/Tiansve/arxiv-cs-subfield)

The model is a frozen sentence-embedding feature extractor plus a small
scikit-learn classifier (linear-probe style). It's a coursework project, not
a production system — predictions near class boundaries are unreliable.