Research On Patent Value Classification Prediction Model Based On Machine Learning

With the advent of the knowledge age,the volume of patent data has exploded,and patent information has become increasingly important.However,the value of patents is uneven,and a method is urgently needed to distinguish this.Traditional patent value evaluation methods require a large number of market indicators,which can no longer adapt to large-scale patent data.The use of machine learning methods can not only reduce the arduous analysis tasks of evaluators,but also improve the efficiency and accuracy of analysis.However,most studies that use machine learning to evaluate patent value only consider better processed structured data in patent samples,and do not establish a connection between patent text and patent value.This research aims to build a comprehensive and scientific patent value index system,and build a predictive model through a series of machine learning algorithms to achieve an objective,accurate and efficient evaluation of patent value.This article first analyzes patent value indicators in depth,uses patent data on public data sets to construct 15 structured indicators containing numerical and sub-types,and 2 unstructured indicators containing textual information,and uses natural language processing methods to integrate The text information is constructed into 5 numerical features that can be used for classification algorithm analysis.Then,the effectiveness of the improved Relief-NMI feature selection algorithm is verified by comparison,and the algorithm is used to rank the importance of the structured index and the unstructured index respectively.Finally,use support vector machine,BP neural network and naive Bayes classification algorithm to analyze the patent data,determine the optimal feature index set,and perform ten-fold cross-validation on the algorithm based on this index set,analyze the performance of the algorithm,and finally get Index system and classification algorithm of patent value classification prediction model based on machine learning.The experimental results show that the accuracy of the classification algorithm after the patent feature selection is effectively improved by the improved Relief-NMI algorithm;the accuracy of the algorithm is further improved after the numerical features composed of text information are added.The optimal feature set is: number of patent citations,type of applicants,number of IPC classifications,legal status,average cosine similarity between the target patent and patents in the same category before the current year,duration of authorization,nationality of applicants,number of claims 8 features;and verified that the support vector machine is higher than the other two algorithms in terms of algorithm accuracy and stability,and the accuracy rate reaches about 90%,and it is a classification algorithm suitable for patent value prediction.