Research On Innovation Diffusion Model Based On Symbolic Regression

Nowadays,the world is in the knowledge economy era,new technologies play an increasingly important role in the company's competition.In order to better assist decision makers to make decisions,more and more attention has been paid to the research of innovation diffusion model.The traditional modeling approaches study the technology diffusion by several critical hypotheses,such as assuming the symmetry as well as structure of model.The experience and judgement of domain experts play vital roles.Although the effectiveness of the hypothetical model could be justified by historical data,it remains a risky task to choose the proper form of the model because erroneous judgements could be made.An intelligent data driven approach is proposed in this paper to learn both structures and parameters of the technology diffusion simultaneously and automatically from historical data,which does not rely on the hypothesis and overcomes the limitation of the existing modeling approach.Based on the empirical study on the historical application data of 95 patents,we will research the performance of the models in 95 technologies.Then we study the diffusion characteristics of the technologies in different categories and countries.And the conclusions we found are as follows:(1)Our approach demonstrates reliable ability for model building.Among the best models returned by our approach,several of the classic models such as the Bass model,the Floyd model and the Fisher model,have been discovered automatically.And the Fisher model and the Bass Model have a wide range of applications.(2)We classify the model according to the characteristics of the model,and we find that the diffusion of most technologies follow the asymmetric models.The asymmetric models whose points of inflection are in between 0.5M and M are the best performing models in our data.(3)By investigating the models discovered in different categories of patents,it could also be observed that the performance of one model is not universally better than that of another.It is suggested that the suitable model should be customized by the exact data where the model will be applied,which the unique advantage is provided by our data driven approach.(4)Different countries have different characteristics of technology diffusion.Different models have different performance in different countries.In summary,symbolic regression method really can find models from data automatically.Symbolic regression can learn the diffusion models from the patent application data.So we can study the performances of the existing diffusion models and research the technology diffusion process in different categories and countries.