基于Lattice LSTM多层次信息交互的装备故障短文本匹配研究

Research on Short Text Matching for Equipment Retrieval Based on Lattice LSTM with Multi-level Information Interaction

为了提高装备检修数据库的检索效率,提出了一种适用于中文句子对的短文本匹配模型Lattice LSTM-Mul。首先,模型采用Lattice LSTM网络获取单词层面的语义特征,避免了中文句子的分词错误问题;接着,采用BiLSTM网络对上下文关系进行建模,更好地捕捉双向语义依赖关系;最后,利用Transformer编码器实现两个短文本的多层次信息交互。对比试验表明,该模型能够提高数据库信息检索功能的用户体验,对推进智慧营区建设起到积极的作用。

In order to improve the retrieval efficiency of equipment maintenance database,a short text matching model named Lattice LSTM-Mul suitable for Chinese sentence pairs is proposed.Firstly,the model uses Lattice LSTM network to obtain semantic features at the word level,which avoids word segmentation errors in Chinese sentences.Then,the BiLSTM network is used to model the context relationship to better capture the bidirectional semantic dependencies.Finally,Transformer encoder is used to realize multi-level information interaction between two short texts.The comparative experiment shows that the model can improve the user experience of database information retrieval function,and play a positive role in promoting the construction of army’s smart barracks.