基于biLSTM的新型文本相似度计算模型

New Text Similarity Computing Model Based on biLSTM

在问答系统中,针对传统神经网络模型在语义匹配准确度不高的问题,提出一种基于双向长短时记忆网络(biLSTM)的快速堆叠编码(SSE)结构融合数据挖掘方法的文本相似度计算模型。该模型先以迁移学习思想将相似度计算的文本量化,分别利用余弦距离和编辑距离计算文本相似度的方法和基于biLSTM结构的三层SSE编码器结构提取文本特征,再将两者提取的特征融合作为最终文本相似度计算特征。实验结果表明,以数据挖掘方法结合SSE模型的F1值高于传统神经网络结构模型。

In question answering system,a text similarity calculation model based on bi-directional long-short memory network(biLSTM)fast stack coding(SSE)structure fusion data mining method is proposed to solve the problem of low semantic matching accuracy of traditional neural network model.In this model,the text is quantized by the transfer learning method,and the text features are extracted by the cosine distance and the editing distance respectively,and the three-layer SSE encoder structure based on LSTM structure,then the features extracted from the two are combined as the final text similarity calculation feature.The experimental results show that the F1 value of data mining method combined with SSE model is higher than that of traditional neural network structure model.