A Patent Quality Evaluation Method Based on Heterogeneous Graph Network

The automated assessment of patent quality plays an important role in promoting technological innovation and enhancing market competitiveness. However, the existing methods cannot fully utilize the multi-dimensional features of patents, leading to low accuracy in quality assessment. In response to this problem, this paper proposes a patent quality assessment method based on heterogeneous graph networks. First, patent numerical attributes and text topics are used to construct a patent heterogeneous graph, and the meta-path is used to sample the heterogeneous subgraph into isomorphic subgraphs under various relationship views. Second, graph convolutional networks are used to perform graph embedding on the heterogeneous subgraphs to obtain patent features. Finally, the attention mechanism is used to fuse the features of all subgraphs to obtain the final representation of the patent node and conduct patent evaluation. The experimental results on two different-sized datasets extracted from the Incopat patent database show that the accuracy of the proposed method is higher than that of the previous methods, proving the feasibility and accuracy of its patent quality assessment.