| Short text is text that is relatively short in length relative to documents and long text.Depending on the granularity of the text,it can be divided into words,sentences or paragraphs.With the spread of the network,users can easily ask questions,express opinions and make comments on the Internet and mobile terminals,and generate a large amount of text information,which is mainly in the form of short text.However,due to the diversity of short text representation methods,the irregularity of words or syntactic structures,traditional algorithms often have problems such as sparse feature vector representation,ambiguity and semantic loss,which don't perform well with tasks related to short text semantic understanding.This paper conducts a study of semantic understanding based on deep learning.It mainly uses deep learning techniques and related algorithms to study the semantic matching and reading comprehension tasks based on semantic understanding.The main contributions and innovations are as follows:1.A model based on dual-layer attention mechanism for semantic matching is proposed.Firstly,an attention-based preprocessing method is used on the word representation layer to reduce redundant information.Secondly,a bilateral multiple perspectives attention mechanism is utilized on the context representation layer to obtain more interactive information.Finally,the obtained information is passed through a Bi-directional Long Short Term Memory Network(Bi-LSTM).Then the obtained final time steps of the two sequences are combined for prediction.The experimental results show that the accuracy of the proposed model in the experimental datasets outperforms the existing advanced benchmark models.2.A model based on the global and local attention interaction mechanism for semantic understanding is proposed.1)The question description is coded by different bidirectional Gated Recurrent Unit(GRU)encoders,and the global and local features of the question information are obtained,then matched with the encoded document information to calculate the semantic similarity under different granularity,which is used to identify document vocabularies of different importance.2)The Avg feature fusion algorithm is used to weight and fuse the obtained global and local feature information,to achieve accurate semantic matching between the document and the problem description.The experimental results show that the proposed model outperforms existing advanced benchmark models on the Children's Book Test(CBTest)dataset.3.A model based on depth separable convolution residual block for reading understanding is proposed.Firstly,a convolution residual block is designed.This module can increasing the network depth while keeping fewer parameters and improve computation efficiency.Then,using the designed module to improve the reading understanding model to achieve more efficient semantic understanding.The experimental results show that compared with the benchmark models,the proposed model further improves the efficiency of model training and reasoning while ensuring the accuracy of the answer inference.In summary,the paper mainly uses deep learning techniques to study the short text semantic understanding and related tasks of different granularities. |
