This repository is the accompanying repository for our paper "What Makes a popular Academic AI Repository".

This paper has been published in Springer's Empirical Software Engineering(CCF-B, SCI journal).

[arxiv] [published version]

Many AI researchers use GitHub repositories to share data, software code and other resources that accompany their publications. We refer to such GitHub repositories as academic AI repositories. In our paper, we collect 1,149 academic AI repositories that are published by AI researchers. We use number of stars as the proxy for measuring popularity of academic AI repositories. Our analysis shows that highly cited papers are more likely to have academic AI repositories with more stars. Hence, in our paper, we perform an empirical analysis of the collected academic AI repositories aiming to highlight good software engineering practices of popular academic AI repositories for AI researchers.

• Python 3
• R (>=3.5)
• Several Python packages including sqlalchemy, gensim, nltk, Beautiful Soup 4, etc.

In the code_repos directory, we provide all the academic AI repositories studied in our paper. Input the following commands to retrieve all the repositories:

git clone https://github.com/YuanruiZJU/academic-ai-repos.git
git submodule init
git submodule update

1. Set the root_path of in the R file: R code/analyze.R.

2. Move all_features.csv to the root_path that is set in Step 1.

3. Run analysis.R in R Commander or RStudio.

According to our experimental results, we have the following checklist for AI researchers when publishing their academic AI repositories: