Research on Disruptive Patent Prediction Based on Outlier Perspective
[Purpose/Significance]In the fierce international competition,disruptive technology is an import-ant opportunity to lead the development of technology and industry,and help enterprises and industries achieve"overtaking on curves".Therefore,to predict and deploy it is of great strategic significance for the country to seize the height of technology and reshape the value chain.[Method/Process]This paper used deep learning and outlier detection algorithms to construct a disruptive patent prediction framework.This research framework included five key steps.Firstly,it adopted the BERT model and TF-IDF algorithm to convert patent text and classification number into computable high-dimensional vector,and took the dimensionality reduction and feature fusion with the PCA algorithm.Secondly,it used three outlier detection algorithms to identify outlier patents through incremental itera-tion.Furthermore,by modifying the dataset,it retained new technology patents from outlier patents.On this basis,by deeply analyzing its core characteristics of disruptive patents in the form of new technologies,it constructed an effective measurement index system.Finally,it took the deep learning DNN model to fit the correlation between patent indicators and disruptive patent labels to effectively predict potential disruptive patents from a lot of new technology patents.[Result/Conclusion]Taking artificial intelligence as an example,it verifies the effectiveness of this method.It predicts 411 disruptive patents in six disruptive directions:multimodal pre-trained large models,augmented reality,generative AI,autonomous driving,image recognition and processing,and intelligent communi-cation,which will have a significant impact on future technological and industrial development.The research results provide important decision-making references for national policy formulation and enterprise technology layout.
NETL
NSTL
万方数据
