基于机器学习分类算法的高质量专利成果筛选研究OACHSSCDCSTPCD

[Purpose/Significance]Based on objective data,the study forms a set of automatic screening methods to quickly identify the quality of patent results and provides decision support to promote the transformation of patent results.[Methodology/Process]Firstly,the study constructed a high-quality patent results screening index system with combining the formal features such as the number of inventors and the number of IPC numbers of patent results with the semantic vector matching degree features and the quality annotation results of patent results;Secondly,taking the field of"advanced man-ufacturing and automation"as an example,the study retrieved the invention patents in this field on the Patent Star platform as the source of patent text data,and took the demand of Hubei Province as an example,and took its relevant industrial development plan(macro)and market technology demand(micro)as the source of demand text data.;then,processed the patented text and the demanded text by using word separation,de-stopping,text vectorization and other steps,and or-ganized to form a training set and a test set;finally,called eight machine learning classification algorithm model for train-ing and evaluation,and tested the algorithm with the best training effect for application to verify the feasibility of the screen-ing method.[Results/Conclusion]The results show that the random forest algorithm model has the best overall perform-ance among the selected eight types of algorithm models,and is used as the kernel classification algorithm in the screening method of high-quality patent results.In addition,the screening method proposed in this paper has a strong feasibility for the quality identification of patent results and can combine the specific patent needs of different provinces(municipalities)to quickly screen large quantities of patent results,which face to a certain extent,effectively reduce the consumption of hu-man,material and financial resources costs.