A novel sentence similarity model with word embedding based on convolutional neural network

摘要

In this paper, we propose an effective model for the similarity metrics of English sentences. Inthe model, we first make use of word embedding and convolutional neural network (CNN) toproduce a sentence vector and then leverage the information of the sentence vector pair to calculatethe score of sentence similarity. Considering the case of long-range semantic dependenciesbetween words, we propose a novel method transforming word embeddings to construct thethree-dimensional sentence feature tensor. In addition,weincorporate the k-max pooling into theconvolutional neural network to adapt to variable lengths of input sentences. The proposed modelrequires no external resource such asWordNet and parse tree. Meanwhile, it consumes very littletime for training. Finally, we carried out extensive simulations to evaluate the performance ofour model compared with other state-of-the-art works. Experimental results on SemEval 2014task (SICK test corpus) indicated that our model can achieve a good performance in the terms ofPearson correlation coefficient, Spearman correlation coefficient, and mean squared errors. Furthermore,experimental resultsonMicrosoft research paraphrase identification (MSRP)indicatedthat our model can achieve an excellent performance in the terms of F1 and Accuracy.