Reproduce “Efficient Patent Searching Using Graph Transformers” 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
Krzysztof Daniell, Igor Buzhinsky, Sebastian Björkqvist
This work introduces a Graph Transformer-based dense retrieval method for patent searching that transforms patent documents into structured graph representations. The approach achieves superior recall for novelty-destroying prior art on a large dataset, outperforming text embedding models and traditional methods, while also enhancing computational efficiency.
Key insights
- Representing patent documents as structured graphs effectively condenses information, leading to more efficient processing compared to using raw, lengthy text.
- Sparse attention mechanisms within Graph Transformers can capture essential relationships in invention graphs without incurring prohibitive computational costs.
- Training with expert-annotated prior art citations enables the model to learn domain-specific relevance, mimicking the reasoning of human patent examiners beyond simple keyword matching.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.