Abstract:

Computing the semantic similarity between two sentences is an important research issue in natural language processing field,and,constructing an effective semantic model of sentences is the core task of natural language processing for paraphrase identification,textual similarity computation,question /answer and textual entailment.In this paper,a parallel convolutional neural network model is proposed to represent sentences with fixedlength vectors,and a similarity layer is used to measure the similarity of sentence pairs.Then,two tasks,namely paraphrase identification and textual similarity test,are used to evaluate the performance of the proposed model.Experimental results show that the proposed model can capture sentence s semantic information effectively; and that,in comparison with the state-of-the-art baseline,the proposed model improves the F1-score in paraphrase identification by 7. 4 percentage points,while in comparison with the logistic regression method,it improves the Pearson correlation coefficient in semantic similarity by 7. 1 percentage points.

Key words: convolutional network, paraphrase identification, sentence model, semantic similarity