A study on patent term prediction by survival time analysis using neural hazard model

Abstract

Patent term is considered one of the factors that determine the private value of a patent. Predicting it can therefore be used as an indicator of corporate management. However, since ordinary regression analysis methods use time series data as the objective variable, it was common to apply the survival time analysis such as the Cox proportional hazards model (CPH) for patent term prediction. On the other hand, CPH cannot incorporate the nonlinear elements of the explanatory variables in the estimation of the risk function, and there is a risk that it may be too simple as a model to predict patent terms from each explanatory variable.Therefore, in this study, we applied Deepsurv, a neural hazard model, to create a prediction model of patent terms for Japanese patent data.As a result, our method improved the prediction performance in terms of Root Mean Squared Error (RMSE) compared to the CPH in the survival time analysis. In addition, it was found that the most important explanatory variable for the prediction performance was the early filing of a request for examination. Furthermore, the RMSE results indicated that the present model may be more suitable for predicting longer patent terms.