A Patent Text Classification Method Based on Phrase-Context Fusion Feature
【摘要】：The traditional patent classification depends on manual classification,which needs the staff to have a strong professional background and has low efficiency. The accuracy of classification is also greatly affected by the high similarity between the subdivision categories. Therefore, automatic classification of patents is needed to speed up the patent classification and improve the accuracy. In this paper, we present a phrase-context fusion feature(PCFF) method for patent classification. The convolutional neural network(CNN) extracts the phrase feature of the patent text, and the bi-directional long short-term memory(BiLSTM) units produce the contextual features. Furthermore, an attention mechanism is developed to adaptively select the contextual features and combines the high separable parts of the features extracted. PCFF model combines the advantages of the two networks and complements each other through fusing the feature from two networks through fusion attention. Experimental results demonstrate that the PCFF model produces complementary fusion feature and obtains better performance against the state-of-the-art baselines on the patent dataset.