Ercnn: Enhanced Recurrent Convolutional Neural Networks For Learning Sentence Similarity

Learning the similarity between sentences is made difficult by the fact that two sentences which are semantically related may not contain any words in common limited to the length. Recently, there have been a variety kind of deep learning models which are used to solve the sentence similarity problem. In this paper we propose a new model which utilizes enhanced recurrent convolutional neural network (ERCNN) to capture more fine-grained features and the interactive effects of keypoints in two sentences to learn sentence similarity. With less computational complexity, our model yields state-of-the-art improvement compared with other baseline models in paraphrase identification task on the Ant Financial competition dataset.