This repository provides a Jupyter Notebook for crawling and visualizing data from the ICLR 2021 OpenReview. It aims to help researchers and practitioners analyze trends, popular topics, and the relationship between keywords and paper ratings within the ICLR conference. The project offers insights into the ICLR 2021 submission landscape, enabling users to identify potentially impactful research areas.

How It Works

The project utilizes Python with libraries like selenium for web scraping OpenReview pages, pandas for data manipulation, and seaborn/wordcloud/imageio/tqdm for visualization. It crawls paper lists and review ratings, then generates visualizations such as keyword frequency word clouds, rating distributions, and analyses of keywords versus ratings to identify trends and potentially successful research themes.

Quick Start & Requirements

• Installation: Requires Python 3.7+, selenium, pandas, seaborn, imageio, wordcloud, and tqdm.
• Webdriver: Requires edgewebdriver or chromedriver (specify path if not in PATH).
• Crawling: Run crawl_paperlist.py (~0.5h) and crawl_reviews.py (~1.5h).

Highlighted Details

• Analyzes 2966 ICLR 2021 submissions.
• Visualizes top keywords (e.g., deep learning, reinforcement learning, GNNs).
• Shows reviewer rating distribution (mean rating ~5.37).
• Suggests keywords like "deep generative models" or "normalizing flows" for potentially higher ratings.

Maintenance & Community

No specific information on maintenance or community channels is provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license.

Limitations & Caveats

The README notes that only review ratings are currently crawled. The crawling process can be time-consuming.