【摘要】 为解决基于Siamese模型缺乏句子间交互的问题和基于匹配模型的匹配因子单一问题,提出双向字粒度交互的卷积神经网络模型。在输入侧通过建立句子交互序列改善Siamese模型交互问题;在特征提取侧和输出侧通过对交互序列进行卷积,建立动态匹配因子改善匹配模型的匹配因子单一问题。实验结果表明,该模型在语义相似性计算数据集Quora和自然语言推理数据集SNLI的准确度相较其它算法均有提升,验证了算法的有效性和可行性。更多还原

【Abstract】 To solve the two problems that the Siamese model lacks the inter-sentence interaction and the matching model lacks enough matching factors,a convolutional neural network model with two-way word-grain interaction was proposed.The Siamese model interaction problem was lessened by establishing a sentence interaction sequence on the input side.By convolving the interaction sequence on the feature extraction side and the output side,dynamic matching factor was established to lessen the matching factor single problem of the matching model.Experimental results show that the accuracy of the model in the semantic similarity computing data set Quora and the natural language inference data set SNLI is improved compared with other algorithms,which verifies the effectiveness and feasibility of the proposed algorithm.更多还原