Syntactic anchoring for artificial intelligence patent insight: A lightweight framework for keyword extraction

Abstract

Compact yet powerful, patent titles embed signals that uncover emerging technological trends. This study introduces a lightweight, syntax-aware method for keyword extraction that identifies functionally meaningful trigrams by leveraging high-frequency prepositions (such as for, on, and using) as structural anchors. Unlike conventional approaches that disregard such function words, the proposed method treats them as semantic pivots, or anchor points in the sentence structure, to capture context-specific expressions, especially in short texts such as patent titles. Applied to 21,100 AI patent titles (2014–2024), the method outperformed six baselines in terms of semantic cohesion (PMI = 11.47), and runtime efficiency, while also demonstrating external validity through alignment with official CPC classification trends (r = 0.73). These results demonstrate the effectiveness of syntactic cues for metadata-level text analysis and highlight the method's practical utility for innovation tracking, patent analytics, and early-stage technology scouting. The study also contributes to the broader discourse on function-oriented innovation by offering a scalable tool for identifying evolving functional expressions in patent corpora.