Research On Potential High Value Patent Prediction Based On Random Forest Algorithm

In the market environment with increasingly fierce competition in science and technology,the earlier the high-value patents are identified and predicted,the greater the space and opportunity for cultivation of high-value patents will be,and the greater the effect and effect will be.In order to support the development of high-value patent cultivation and meet the demand of potential high-value patent prediction,this paper proposes a potential high-value patent prediction method based on random forest algorithm to realize the purpose of predicting potential high-value patents in the application and examination stages.The prediction of high-value patents in the early stage of patent life cycle--application and examination stage,on the one hand,is conducive to the targeted allocation of limited social resources for the cultivation of high-value patents with potential,so as to further promote and realize the value of highvalue patents.On the other hand,shortening the time difference between the application,implementation and transformation of high-value patents can help Chinese enterprises grasp the opportunity and opportunity in the international market competition,and seize the market and seize the initiative in the early stage of technology development.On the basis of fully combing the existing research,the index system of potential high-value patent prediction is constructed by comprehensively considering the legal factors,technical factors and market factors that affect the patent value.In order to overcome the problems of late identification stage and high labor cost,and considering that the prediction of potential high-value patents needs to face the mass of early patent applications,high timeliness,and the method needs to be explicable and easy to be popularized,a method of potential high-value patents prediction based on random forest algorithm is proposed.In the experiment,the BINGO method was firstly used to merge the patent value data of Innography,Incopat and Pat Snap patent information platform.It was verified by experts that the BINGO fusion method effectively improved the quality of experimental data.Then,a random forest prediction model is constructed on the potential high-value patent prediction index system,the prediction results are tested,and a comparative experiment is designed to prove the effectiveness of the model.Finally,19,647 patents in the field of "speech signal recognition" were used for empirical analysis,and the prediction accuracy of the model reached 96%.The method proposed in this paper makes up for the deficiency that is currently applicable to the research of discovering potential high-value patents from the mass of early applications.It can predict potential high-value patents in the stage of patent application and examination,and has the advantages of high classification accuracy,good prediction timeliness,large amount of data processing and good model interpretability.Further research can try to improve the robustness of the method or discuss the applicability of the method in more patent technology fields.