In the process of high value patents evaluation, the important features of patents obtained through feature selection largely affect the performance of the classifier. The traditional feature selection method has the problem of single evaluation factor, and should combine multiple evaluation factors for a more comprehensive screening of features, so this paper proposes a feature selection of high value patents based on random forest, which obtains through the random forest algorithm The method obtains the influence factor, enhancement factor and importance factor of features by random forest algorithm, and combines the three factors for a more comprehensive screening of the original patent feature set to obtain the important features of high value patents. Through comparison experiments with the single factor method, it is proved that the method has better feature selection effect.