PulseAugur
EN
LIVE 08:19:46

New research explores LLM-based patent analysis and benchmarking

By PulseAugur Editorial · [4 sources] ·

Two new research papers explore methods for improving patent representation learning using large language models. The first paper introduces PHAGE, a novel encoder that uses heterogeneous dependency graphs to better capture the hierarchical structure of patent claims, outperforming existing baselines in classification, retrieval, and clustering. The second paper benchmarks 22 different embedding models, evaluating their performance across retrieval, classification, and clustering tasks, and finds that fine-tuning strategies are task-dependent and that single-landscape fine-tuning can degrade performance on external landscapes. AI

IMPACT These studies highlight advancements in applying LLMs to specialized domains like patent analysis, suggesting improved efficiency and accuracy in information retrieval and classification.

RANK_REASON The cluster contains two academic papers detailing novel methods and benchmarks for patent representation learning using LLMs.

Read on arXiv cs.IR (Information Retrieval) →

AI-generated summary · Google Gemini · from 4 sources. How we write summaries →

New research explores LLM-based patent analysis and benchmarking

COVERAGE [4]

  1. arXiv cs.CL TIER_1 English(EN) · Yongmin Yoo, Qiongkai Xu, Zhangkai Wu, Longbing Cao ·

    Heterogeneous Dependency Graph-Guided Attentionfor Patent Representation Learning

    arXiv:2605.10073v2 Announce Type: replace Abstract: Pre-trained language models advance patent classification and retrieval via encoding claims as flat token sequences, yet overlooking the dependency hierarchy among claims. Incorporating the hierarchy into self-attention poses tw…

  2. arXiv cs.AI TIER_1 English(EN) · Amirhossein Yousefiramandi, Ciaran Cooney ·

    Benchmarking Patent Embeddings: A Multi-Task Evaluation of 22 Models Across Retrieval, Classification, and Clustering

    arXiv:2605.24297v1 Announce Type: cross Abstract: Which fine-tuning signals improve patent embedding models, and do gains transfer across patent landscapes? We benchmark 22 embedding models, from 22M-parameter encoders to 12B instruction-tuned LLMs, on retrieval, classification, …

  3. arXiv cs.IR (Information Retrieval) TIER_1 English(EN) · Ciaran Cooney ·

    Benchmarking Patent Embeddings: A Multi-Task Evaluation of 22 Models Across Retrieval, Classification, and Clustering

    Which fine-tuning signals improve patent embedding models, and do gains transfer across patent landscapes? We benchmark 22 embedding models, from 22M-parameter encoders to 12B instruction-tuned LLMs, on retrieval, classification, and clustering. The study uses 113,148 WIPO assist…

  4. arXiv cs.IR (Information Retrieval) TIER_1 English(EN) · Ciaran Cooney ·

    Benchmarking Patent Embeddings: A Multi-Task Evaluation of 22 Models Across Retrieval, Classification, and Clustering

    Two questions regarding practitioners' use of patent embeddings arise: (i) Does one fine-tuning recipe suffice for all downstream applications? (ii) Is fine-tuning on one patent landscape sufficient for downstream application on other landscapes? By evaluating 22 pre-trained embe…

RELATED ENTITIES

RELATED TOPICS