Prediction of High-value Patents by Incorporating Knowledge Network Embeddedness

Accurate prediction of patent value and early identification of patents with high value are of great significance to promote the cultivation of high-value patents and technical layout. Based on knowledge reorganization and patent invention creation process, the article selects and designs indicators of the knowledge network embeddedness to represent the association between sample patents' knowledge and the prior knowledge of their domain. By having integrated the characteristics of innovation actors and patent application, a variety of machine learning models are built to predict the value of the patents at early stage of their application. The high-value patents in the field of neural networks are studied empirically, and the F1 value of the proposed high-value patent prediction model reaches 80%, and the prediction results are effective. Meanwhile, knowledge network embeddedness (especially PageRank and eigenvector centrality) plays an important role in predicting high-value patents.