Optimization of Machine Learning-based Value Assessment Model in High Value Patent Cultivation in Universities

Abstract

This paper constructs a patent value assessment model for colleges and universities from two perspectives of: value identification and price prediction. Firstly, 10 indicators are selected from 3 dimensions of technology, economy, and law. Then it combines the artificial way of entropy weight TOPSIS model and the machine learning way of gradient boosting tree to realize the identification of the value of university patents and the grading of the economic value of university patents. After analyzing, it can be seen that after pre-processing the data, 10 feature items related to patent value and useful for evaluation are screened out, and the highest weights of the number of homologous patents and the number of citations indicators are 0.1826 and 0.1274, respectively, which have the greatest influence on the economic value of high-value patents of colleges and universities. In the range of 4901-7071 of high-value patents, the assessment results fluctuated in the range of 1.3754-2.8395. The value of invention patents with a gradient range of 1-1400 as well as 6301-7071 fluctuates more dramatically. This paper proposes a patent value assessment model for universities that has a superior assessment and classification effect on high-value patents in universities.

Keywords

• entropy weight topsis
• gradient boosting tree
• assessment model
• university patent value
• 68m11