A context-aware framework for strategic patent intelligence: keyword expansion and non-patent literature classification

Purpose Most patent search systems rely on simple similarity measures and provide limited contextual understanding or integration with non-patent literature (NPL). This study proposes a context-aware patent search system (CAPSS) that improves keyword recommendation and cooperative patent classification (CPC)–based NPL classification within a unified framework to support strategic patent intelligence and early trend detection. Design/methodology/approach CAPSS integrates patents and NPL through context-aware retrieval. Domain-specific keywords were expanded using FastText to capture word-level variations, and BERT was used to refine the results by considering sentence-level context. NPL was classified into CPC codes using fine-tuned transformer models (BERT, RoBERTa, DistilBERT and PatentBERT) with discriminative, gradual unfreezing and hybrid fine-tuning strategies. Findings CAPSS outperformed conventional keyword-based approaches in search accuracy and analytical efficiency. FastText improved domain-specific term expansion, BERT improved contextual relevance and the BERT-Hybrid model achieved the best NPL classification performance. CPC-based labeling further supported integrated patent-NPL analysis. Originality/value This study presents the first integrated retrieval system that combines FastText–BERT for keyword expansion and CPC-based linking of patent and NPL. The system enhances both the accuracy and efficiency of prior art searches and proves useful for analyzing technology trends, patent review and R&D planning.