Interactive Context-Comparative Model for Text Matching

Most of the existing text matching models are based on the interactive network and use cross-attention mechanism to capture the semantic links between two texts, and have made outstanding achievements. However, the current methods generally do not pay enough attention to the context information of the text and ignore the importance of comparing these features. To solve this problem, this paper proposes an interactive context-comparative model for text matching (ICCM). First, the Bi-directional Long Short-Term Memory network is used to encode the word vector. Then the key information of related words between different texts is aligned through the cross-attention mechanism, and the key information in their contexts is focused through the context-attention mechanism. The two kinds of information are compared using comparison functions, and the convolutional neural network is used to aggregate the features after splicing. Finally, the results of text matching are predicted. Testing on LCQMC dataset shows that the accuracy rate is 85.5%. The experimental result and analysis prove that the proposed model is superior to the existing main text matching methods and has certain effectiveness.