Automatic classification method for new energy patents based on improved DPCNN

Inhalt:
Patent text has very important research value, and the traditional patent classification method is generally based on the IPC classification standard. the IPC classification standard is relatively coarse, and with the rapid growth of the number of patents, the standard can no longer meet the real situation, and a patent classification method is urgently needed to help various industries achieve accurate classification of patents. To this end, this paper proposes an improved patent classification model of DPCNN. First, a Concat layer is added to DPCNN to fuse the features of each layer of the network and improve the problem that traditional neural networks tend to lose structural information. Then, a label smoothing strategy is introduced to avoid overconfidence of the neural network in training answers during the training process. Finally, the effectiveness of the model is verified by taking a new energy patent as an example.