关键词: 问题相似度, 递归卷积神经网络, 全局最大池化, 孪生网络

Abstract: Using deep learning to calculate the question similarity in search engine and question answering system is a hotspot in the NLP field. Combining convolutional neural network (CNN) and long-short memory network (LSTM), a recursive convolutional neural network (RCNN) question similarity calculation method is proposed. Firstly, the bidirectional recursive neural network is utilized to extract context information, and then the 1D Convolutional Neural Network was used to integrate the word embedded information with the context information. Then the global maximum pooling is used to extract the key information to complete the semantic representation of the two questions.Finally, the similarity of the question pair is judged through the matching layer. The experimental results show that, based on the Quora Question Pairs data set, the accuracy of the question similarity calculation method is 83.57%, which is better than other methods.

Key words: question similarity, recursive convolutional neural network, global maximum pooling, siamese network