Reproduce “Adaptive Multi-Stage Patent Claim Generation with Unified Quality Assessment” 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

Chen-Wei Liang, Bin Guo, Zhen-Yuan Wei, Mu-Jiang-Shan Wang

An adaptive multi-stage framework automates patent claim generation and quality assessment, achieving a ROUGE-L score of 52.8 on USPTO and 54.3 on EPO datasets. The system also demonstrates an 0.847 correlation with human expert judgment for quality evaluation and 15% faster training convergence.

Key insights

  • Explicit modeling of diverse patent relationships through specialized multi-head attention effectively captures nuanced semantic dependencies critical for patent analysis.
  • Combining dynamic LoRA adapter selection with curriculum learning enhances an LLM's adaptability to various legal jurisdictions and technical domains.
  • A unified multi-task quality assessment framework, leveraging cross-attention and adaptive margin learning, significantly improves the consistency and human alignment of automated evaluations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.