DeepGit: Promoting Exploration and Discovery of Research Software with Human-Curated Graphs

Yilin Xia, Shin-Rong Tsai, Matthew Turk

Keywords: Research Software Discovery, Graph-based Exploration, GraphRAG

Abstract: Familiarizing oneself with a new research field increasingly demands not only reading academic literature, but also exploring domain-specific research software tools. However, unlike academic publications, the software landscape lacks centralized platforms comparable to Google Scholar, making the "tool review" process challenging. Although GitHub offers some support in this space, the software discovery process is often biased by popularity metrics and offers limited insights into relationships between repositories. We present DeepGit, a domain-aware engine designed to support software discovery from GitHub metadata. DeepGit employs a human-in-the-loop methodology, where researchers and domain experts collaboratively guide software discovery by finalizing research topics, constructing graphs to suit research needs, and extracting subgraphs for subsequent question answering. This approach bridges the gap between customized needs and automated exploration, enabling more user-oriented, comprehensive, and interpretable discovery of research software.

Submission Number: 4