Multi-View Sentence Matching Model Based on Equal-Length Interactive Attention

Sentence matching aims to determine whether there is a semantic connection between two sentences. Currently, most sentence matching models align elements in two sentences by stacking multiple layers of attention networks to obtain competitive results. However, this approach has some drawbacks. This paper proposes a multi-view sentence matching model based on equal-length interactive attention. It can use multiple equal-length interaction units to capture more accurate attention information and speed up the model's process of extracting high-level features. Then, we introduce a residual connection mechanism to alleviate the gradient problem of the network. Finally, we design a multi-feature spatial information extraction module to extract the detailed information contained in the underlying features. Experiments demonstrate that our proposed method is effective and achieves competitive results on three datasets, LCQMC, PAWS-ZH, and PAWS-EN.