Research And Application Of Short Text Semantic Similarity Model Based On Deep Learning

The development and application of Natural Language processing(NLP)is an important link in the development of modern science and technology.Short-text semantic similarity calculation is widely used in NLP task,which is the foundation of NLP field,and its importance is self-evident.Therefore,it is one of the important tasks for the development of modern science and technology to vigorously study and improve the calculation method of text semantic similarity.In this paper,the semantic similarity calculation of short text adopts the deep learning method.The deep learning model is used as the sentence encoder to encode the text,and the distance of the encoded one-dimensional vector is calculated.Finally,the distance value obtained is used as the text similarity calculation value.The main work of this paper is as follows:(1)This paper proposes a COV-BIGRU text coder based on Convolutional neural network(CNN)and Gated recurrent unit(GRU).COV-BIGRU encoder extracts important semantic information and text sequence information simultaneously.Overcome the failure of CNN to extract sequence information,also overcome the Recurrent neural network(RNN),Long short-term Memory(LSTM)and other RNN variants that cannot effectively extract important semantic information from text.Finally,the COV-BIGRU text encoder is compared with the commonly used deep learning text encoder,and the Quora data set,MSRP data set and Sick data set are used for verification.The final results show that the F1 value of the COV-BIGRU encoder is the highest in the three data sets.In the Quora dataset,F1 value is 90.02%,8.39% higher than LSTM encoder and 4.2% higher than CNN encoder F1 value.(2)Based on deep learning text similarity calculation model and Elasticsearch search engine technology,a search system based on semantic similarity algorithm is designed and implemented.The system can return semantically similar content to the user's search content in the database when there is no search content in the database.