Research on High-Value Patent Prediction Method Based on Technology Component Mapping Network

This research aims to develop a sophisticated method for precisely predicting high-value patents by integrating a technology component network with patent metric features. This integration is based on the potential semantic correlations between the technical components embedded in patents, thereby providing robust technical support for optimizing technology innovation management. The proposed method initially extracts Subject-Action-Object (SAO) triplets and constructs a technology component network by leveraging the semantics of the Action component. This process maps the relationships between technology components to the relationships among patent documents. Subsequently, patent document metric features are utilized as node attributes, while the mapped patent document relationships serve as edges to form a patent semantic association network. Ultimately, a graph convolutional neural network (GCN) model is employed to predict high-value patents. The experimental results demonstrate that the method based on the patent technology component mapping network (CMN) generally outperforms the prediction method relying on the patent citation network (PCN) and the patent semantic similarity network (PSSN). Specifically, the GraphSAGE-based CMN prediction method achieves the best performance, with an F1 score of 0.81, which represents a notable improvement over the PCN-based method and the PSSN-based method. Additionally, case study analysis further substantiates the effectiveness and superiority of the proposed high-value patent prediction method.