This is the repository of ReView: A Tool for Visualizing and Analyzing Scientific Reviews. Code | Hugging Face Spaces

Installation

• Since this project was built with Python 3.10, first create the following virtual environment:

module load miniconda/3
conda create -n ReView python=3.10

• Second, activate the environment and install pytorch:

conda activate ReView

• In the next step, make sure you have git-lfs package and install git lfs, since it is used in this repo for the preprocessed dataset:

conda install git-lfs
git lfs install

• Finally, all remaining required packages could be installed with the requirements file:

pip install -r requirements.txt

• (Optional) To enable fetching reviews directly from OpenReview links in the Interactive tab:

pip install openreview-py

Running Interface Locally

To run this interface locally, first, make sure gradio and all the requirements are installed in your environment. Then, you can run the following for a local instance of the interface:

python interface/Demo.py

Note: Do not use gradio ./interface/Demo.py (hot-reload mode), as it is currenlty incompatible with the interface's dynamic UI.

Additionally, you can edit the last line of code for a shareable link of your local instance as desired (change demo.launch(share=False) to demo.launch(share=True))

Introduction and Instructions

For an up-to-date, concise, and brief introduction of the interface, you can check out the "Introduction" tab of the ReView app here

Data Processing Pipeline

All data processing scripts live in the pipeline/ directory:

File	Purpose
pipeline/run_scoring.py	Unified end-to-end pipeline orchestrator
pipeline/process_new_data.sh	Shell entry point (forwards to run_scoring.py)
pipeline/fetch_iclr_data.py	Fetch reviews from OpenReview API
pipeline/preprocess_data.py	Text cleaning and preprocessing
pipeline/run_glimpse_scoring.py	GLIMPSE consensuality/agreement scoring
pipeline/run_polarity_scoring.py	Polarity/sentiment scoring
pipeline/run_topic_scoring.py	Topic/aspect scoring
pipeline/scored_reviews_builder.py	Integrate all scores into final dataset
pipeline/config.py	Centralized configuration

The pipeline auto-detects available years from the data/ directory. To process data for any year:

# Fetch data for a new year
python pipeline/fetch_iclr_data.py --year 2026

# Run the full scoring pipeline (auto-detects all available years)
./pipeline/process_new_data.sh

# Or run for a specific year
./pipeline/process_new_data.sh --year 2026

Performance

Since this project was built for deployment on Hugging Face Spaces, it is optimized to run on CPU. However, if better performance is needed, you can run this interface on a CUDA-enabled device and profit from the improved performance of the models in the interactive page. The code is set up to automatically use CUDA if available.