Scholarly Knowledge Extraction from Published Software Packages

TL;DR

This paper presents an automated method for extracting structured scholarly knowledge from published software packages by analyzing metadata, code, and related literature to enhance research reproducibility and knowledge graphs.

Contribution

It introduces a novel approach combining static analysis, metadata extraction, and literature linking to build a scholarly knowledge graph from software packages.

Findings

Successfully extracted metadata and code procedures from software packages.

Linked software information to scholarly articles for contextual understanding.

Published extracted knowledge in the Open Research Knowledge Graph.

Abstract

A plethora of scientific software packages are published in repositories, e.g., Zenodo and figshare. These software packages are crucial for the reproducibility of published research. As an additional route to scholarly knowledge graph construction, we propose an approach for automated extraction of machine actionable (structured) scholarly knowledge from published software packages by static analysis of their (meta)data and contents (in particular scripts in languages such as Python). The approach can be summarized as follows. First, we extract metadata information (software description, programming languages, related references) from software packages by leveraging the Software Metadata Extraction Framework (SOMEF) and the GitHub API. Second, we analyze the extracted metadata to find the research articles associated with the corresponding software repository. Third, for software contained in published packages, we create and analyze the Abstract Syntax Tree (AST) representation to extract information about the procedures performed on data. Fourth, we search the extracted information in the full text of related articles to constrain the extracted information to scholarly knowledge, i.e. information published in the scholarly literature. Finally, we publish the extracted machine actionable scholarly knowledge in the Open Research Knowledge Graph (ORKG).