Abstract:Sentence Similarity Computation lays the basis of many natural language processing tasks, the accuracy of which has a direct impact on the performance of language related systems, especially in machine translation, plagiarism detection, query ranking and question answering. Therefore, it is in demand to raise the accuracy of sentence similarity computation by directing traditional methods that rely on shallow features like morphology, word sequence and grammar structure to employ deep learning methods, or integrating these two approaches together. However, deep learning methods using convolutional neural networks are required to overcome disadvantages such as narrow receptive field and insufficient long-distance information dependence when extracting text features. This paper carries out dependency-based syntactic analysis to retrieve information over longer distance. We make text parsing employing Stanford NLP for syntactic analysis, and then retrieve mutual relationship between two words in a binary combination or triplet. As lexical supplement information embedded in these word combinations, the dependency information, in addition to the original sentence, is added up as Convolutional Neural Network input, thus constructing a Dependency CNN. Experiment results reveal that the long distance dependency information effectively improve the similarity computation performance in our proposed dependency model on MSRP dataset, reaching an accuracy and F1 value of 80.33% and 85.91 respectively. Notably, Pearson correlation coefficient of this long distance dependency model reaches 87.5 on SICK dataset, and a much higher 92.2 on MSRvid dataset.