Research on acrostic poem generation based on handwritten Chinese character recognition and machine learning

Abstract

In this paper, we study the automatic generation of acrostic poems based on handwritten Chinese character recognition and machine learning. First, AlexNet network and ResNet50 network are used to train the handwritten Chinese character data set respectively. Select ResNet50 which recognition accuracy is 82.56% as the input module of poem generation model. After this, the GPT-2 model is used to train 26,000 five-character poems in order to build an acrostic poem generation model. Finally, combine these two parts. If we input four images of handwritten Chinese characters, the system can automatically generate a fluent acrostic poem headed by these four characters after character recognition. Experiment results verify the effectiveness of the method.