Chinese Patent Approval Prediction Based on Artificial Intelligence

The use of artificial intelligence methods to process patent text and realize automated patent approval helps to assist patent examiners and speed up the approval progress. However, existing research rarely involves the field of patent approval and lacks the support of corresponding Chinese datasets. To solve this problem, this paper proposes a Chinese Patent Approval Prediction Model Based on Artificial Intelligence (AIPat) to improve the prediction performance. The model calculates for each patent to be evaluated its maximum similarity to the prior art, constructs a structural graph based on the reference relationship between the claims, and obtains a structural patent representation by encoding the fused textual and structural information. Starting from the drafting specification of Chinese patent claims, the representation is disentangled into two subspaces of constituent elements and element relationships, constrained by BoW prediction and parent claim prediction respectively. Finally, the disentangled representations are fused with similarity scores for patent approval prediction. To accomplish this task, we constructed three Chinese patent datasets in different domains, and the experiments conducted on them proved the superior performance of the model and provided directions for further research in this area.