---
title: acl-spectrum
emoji: 📊
colorFrom: gray
colorTo: pink
sdk: streamlit
sdk_version: 1.26.0
app_file: app.py
pinned: false
license: mit
---

# Navigating ACL Anthology

With the goal of understanding how scientific papers are related to one another, we plot their projections on a 2D plain in our interactive [demo](https://huggingface.co/spaces/gwf-uwaterloo/acl-spectrum).

## Data
We downloaded meta information of [ACL Anthology](https://aclanthology.org/) papers from [here](https://github.com/acl-org/acl-anthology/tree/master/data/xml).

Overall, 12,282 papers were derived from 4 conferences/journals, i.e. ACL (5.2K), EMNLP (4.6K), NAACL (2K), and TACL (442), over 7 years (2016-2023).

## Embeddings
We encoded the papers (titles + abstracts) using [spectre2](https://huggingface.co/allenai/specter2_base), a BERT-based model designed for scientific literature.
We applied t-SNE to reduce dimensions for visualization.

To learn about how we generated the plot, check out our step-by-step guide [here](scipapers_scatter.ipynb).

👉 Released under the [MIT License](LICENSE).

👉 For questions (❓), problems (🛠️), or bug reports (🐛), please open an [issue](https://huggingface.co/spaces/gwf-uwaterloo/acl-spectrum/discussions) under the Community tab.