[Research purpose]In view of the strong description of patent text,the traditional high-value patent classification adopts the index calculation,and fails to consider the detailed context information or relatively long text sequence in patent documents.This paper proposes a method to identify high-value patents by combining the content of patent text and the feature information of patent indexes.[Research method]It proposes a high-value patent recognition model based on content understanding and index fusion.Firstly,BERT-BiLSTM is used to extract the contextual and sequential features of patent text,and then the extracted features of patent text and patent in-dex are fused,and finally XGBoost algorithm is used to complete the high-value patent classification.[Research conclusion]After sev-eral groups of comparative experiments,the method proposed shows an accuracy of 74.19%in identifying the winning of the China Patent Award granted by the State Intellectual Property Office of China in the field of identifying basic electrical components and electrical com-munication technology,a recall rate of 76.66%,and an F1value of 75.40%.The results show that the method can effectively improve the classification accuracy of high-value patents.
维普
万方数据