[Objective] The method of identifying different types of star inventors by the number of patents and patent citations has obvious time lag. It is a new approach to identify innovation types of star inventors early based on patent text information.

[Methods]This paper classifies the innovation types of star inventors into four categories: “composite”, “consolidation”, “breakthrough” and “development” by taking “continuity innovation” and “breakthrough innovation” as the division dimensions, and constructs a model for identifying star inventor types based on graph convolutional neural network by combining the patent title information and the cooperation relationship of star inventors.

[Results]Experiments with patent data in the field of molecular biology and microbiology show that the overall accuracy of the model in identifying the innovation types of star inventors is 79.4%, which is 15% higher than that of word vector alone.

[Limitations]The identification effect of the model in this paper is not ideal for “breakthrough star inventors”, and further research is needed to find the characteristics of breakthrough inventors to improve the validity of the model.

[Conclusions]This method can overcome the time-lag effect of the identification method based on the number of patents and citation and identify the innovation type of star inventors as soon as possible.