LSGC: An Interactive Text Matching Model Combined with Enhanced Encoding
Chenglin Wang, Xiaowei Xu, Zhimin Wei, Wei Zhang, Zhenyu Liu · 2022 International Joint Conference on Neural Networks (IJCNN) · 2022
Text matching which measures the degree of similarity between text pair plays a vital role in a variety of applications. However, a single process of encoding such as BiLSTM may lose textual information. In this paper, we propose a new interactive text matching model named LSGC, where we design an enhanced encoding layer that uses not only BiLSTM but also gated convolutional neural network to enhance the ability to capture both key word and sequence information. Additionally, for the purpose of avoiding long-term dependency and fusing more important information in the fusion layer, we take advantage of BiLSTM and attention to enhance the inference fusion. Finally, experiment results demonstrate that our model outperforms other neural network models and achieve a remarkable performance on both Chinese and English paraphrase identification datasets.