Reproduce “Analogy Search Engine: Finding Analogies in Cross-Domain Research Papers” 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
Jieli Zhou, Yuntao Zhou, Yi Xu
Analogy Search Engine (ASE) treats scholarly discovery as a structured analogy problem: it retrieves papers with similar research purposes but different methods or mechanisms. Using segmented abstracts from the Solvent 2K AI-paper dataset, ASE combines embedding-based nearest-neighbor search with clustering and achieved roughly 10% higher user-rated usefulness and interestingness than Elasticsearch in a small study of 21 query papers.
Key insights
- Functional segmentation makes it possible to search for specific relationships between papers instead of collapsing every abstract into one undifferentiated similarity score.
- KNN-KMeans was the most stable approach on the small corpus because it preserved purpose similarity while using corpus-wide clustering to remove methodically similar candidates.
- Embedding distance was more reliable for topical or content similarity than for identifying genuinely different mechanisms; large vector distance does not necessarily imply a meaningful conceptual analogy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.