A Python package (and website) to automatically attempt to find GitHub repositories that are similar to academic papers.

Stable Release: pip install papers-without-code
Development Head: pip install git+https://github.com/evamaxfield/papers-without-code.git

Provide a DOI, SemanticScholarID, CorpusID, ArXivID, ACL, or URL from semanticscholar.org, arxiv.org, aclweb.org, acm.org, or biorxiv.org. DOIs can be provided as is. All other IDs should be given with their type, for example: doi:10.18653/v1/2020.acl-main.447 or CorpusID:202558505 or url:https://arxiv.org/abs/2004.07180.

pip install papers-without-code
pwoc query
# or pwoc path/to/file.pdf

from papers_without_code import search_for_repos
search_for_repos("query")
# search_for_repos("path/to/file.pdf")

⚠️ Prior to using PWOC with a PDF you must be logged in to Docker CLI via docker login because we automatically fetch, spin up, and tear down containers for processing. ⚠️

In short, we pass the query on to the Semantic Scholar search API which provides us basic details about the paper. We use a prompted gpt-3.5-turbo with langchain to extract keywords from the title and abstract. We then make multiple threaded requests to GitHub's API for repositories which match the keywords. Once we have all the possible repositories back, we rank them by similarity between the repository's README and the paper's abstract (or if not available, it's title).

When using Papers without Code locally and providing a filepath, the only change to this workflow, is paper details gathering. When local and providing a filepath, we use GROBID to extract the title, abstract, and author list.

For full package documentation please visit evamaxfield.github.io/papers-without-code.

Exploratory data analysis of the dataset used for testing

See CONTRIBUTING.md for information related to developing the code.

MIT License