Is It Novel and Why? Fine-Grained Patent Novelty Prediction Based on Passage Retrieval

2026-05-04 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Intellectual Property & Patents · Depth: Advanced, quick

Summary

A new dataset, FiNE-Patents (Fine-grained Novelty Examination of Patents), has been introduced to improve patent novelty assessment. This dataset contains 3,658 first patent claims, each annotated with feature-level prior art references derived from European Search Opinion (ESOP) documents. The research proposes a shift from traditional claim-level binary classification to a joint retrieval and abstract reasoning task at the feature level. This approach requires models to identify specific passages in prior art that disclose individual claim features and to pinpoint which features contribute to a claim's novelty. LLM-based workflows were implemented and evaluated, demonstrating superior performance over embedding-based baselines in passage retrieval and novel feature identification. These LLM workflows also exhibit robustness against spurious correlations often found in claim-level novelty classification tasks.

Key takeaway

For research scientists developing patent examination tools, you should consider adopting a feature-level analysis paradigm. This approach, supported by the FiNE-Patents dataset and LLM-based workflows, offers more granular insights into novelty and reduces susceptibility to spurious correlations, leading to more accurate and transparent assessments.

Key insights

Fine-grained, feature-level analysis improves patent novelty prediction and mitigates spurious correlations.

Principles

• Novelty assessment requires feature-level granularity.
• LLMs are robust against spurious correlations.

Method

Decompose patent claims into features, analyze each feature against prior art using LLMs, then derive a claim-level novelty prediction.

In practice

• Use FiNE-Patents for granular patent analysis.
• Apply LLMs for feature-level prior art matching.

Topics

• Patent Novelty Prediction
• Fine-grained Patent Analysis
• Passage Retrieval
• LLM Workflows
• FiNE-Patents Dataset

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Legal Professional

Related on AIssential

• Building a Semantic Search Engine for Patent Prior Art Discovery Using SBERT and FAISS — Semantic embeddings significantly enhance patent prior art discovery by retrieving documents based on meaning, not just keyword overlap.
• FineREX: Fine-Tuned NER-RE for Human Smuggling Knowledge Graphs — Domain-specific fine-tuning of LLMs substantially improves knowledge graph quality and efficiency for complex legal text.
• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — Existing RAG benchmarks misrepresent real-world performance due to high document redundancy and similarity in enterprise corpora.
• Revisiting the Systematicity in Negation in the Era of In-Context Learning — LLMs partially grasp negation via in-context learning, but struggle with robust scope recognition and function vector composition for it.
• Extracting Article-Level Legal Dependencies from Swiss Federal Law using LLMs — LLMs can automate the extraction of legal dependencies from complex federal law articles.
• NLP-based Page Classification for Efficient LLM Extraction from Brazilian Public Tender Documents — Combining NLP page classification with LLM extraction significantly reduces token count and cost for document processing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.