摘要
Technological innovation is often achieved by combining existing technological elements in novel ways. However, as the number of technological elements increases, the number of possible combinations grows rapidly, making it difficult for firms to identify feasible opportunities for recombination. In particular, systematic analysis considering the functional relationships between technological elements is required. To address this problem, this study proposes a novel approach based on subject-action-object (SAO) structures extracted from patents. The approach represents technological knowledge as a knowledge graph and uses this representation to discover opportunities for technological recombination. First, SAO structures are extracted from patent texts. The subjects, actions, and objects are then generalized using unsupervised machine learning techniques and modeled as an SAO knowledge graph. Subsequently, a triple prediction model is developed that integrates structural relationships in the knowledge graph with semantic information from SAO texts to screen feasible technological combinations. Finally, three paths for technological recombination opportunity discovery are proposed to identify opportunities involving solution components, functions and effects, and target problems from a problem–solution perspective. The applicability of the proposed approach was empirically evaluated using 37 737 patents in the energy storage sector. The prediction model achieved an F1-score, PR-AUC, and MCC of 0.8300, 0.7967, and 0.6600, respectively, outperforming the second-best comparison method by 3.92% in terms of F1-score. Furthermore, a case study of a target firm suggests that the proposed approach can support recombination opportunity screening, engineering design, and industrial R&D planning. The rigor of this study was strengthened through multiple validation procedures, including baseline comparisons, ablation studies, sensitivity analyses, and future-triple prediction.