Reproduce “Matlas: A Semantic Search Engine for Mathematics” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haocheng Ju, Leheng Chen, Peihao Wu, Bryan Dai, Bin Dong
Matlas is a semantic search engine designed for mathematical statements, built on a large corpus of 8.07 million statements extracted from 435,000 peer-reviewed papers and 1,900 textbooks. It features a novel dependency-aware unfolding mechanism to produce more self-contained statement representations, facilitating retrieval and interpretation for both human researchers and AI systems.
Key insights
- Curating a corpus primarily from peer-reviewed journals and textbooks, selected using an ICM citation-based criterion, provides a more reliable and historically deep foundation for mathematical knowledge retrieval.
- Dynamically generating regular expressions and leveraging LLMs in a two-stage process allows for robust extraction of diverse mathematical statements and their local dependencies from complex, varied document structures.
- Recursively incorporating dependencies to unfold mathematical statements into self-contained representations addresses the inherent context-dependency of mathematical language, making them more interpretable for both humans and AI without requiring external context lookups during inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.