Research Of English Textual Entailment Recognition Based On Deep Neural Network

With the advent of the information age,text data on the Internet is growing.Therefore,fast and efficient access to useful information becomes very important.In the field of natural language processing,one of the basic techniques for the "understanding" of computer-text entailment recognition has aroused more and more researchers ’ attention.Textual entailment recognition is the process of recognizing the directional semantic relationship between texts.The main problems of the textual entailment recognition are as follows:There are many synonyms,abbreviations and pronunciations in the English texts,which will affect the result of entailment recognition.Because of complex grammatical structures in English,different expressions sometimes mean the same thing.Deep semantic understanding is also required for English textual entailment recognition.By analyzing the main problems of the textual entailment recognition and the shortcomings of the previous works,we propose an approach based on the deep neural network in this paper.Firstly,preprocess the text.Then,use the Long Short-Term Memory network to extract the semantic information of the texts and align the text softly based on attention mechanism,where a feed forward network is used to compare the text semantically.Finally,a fully-connected neural network is used to obtain the entailment relationship.When the training of neural network is completed,the entailment recognition features of neural network is merged with the traditional linguistic features to re-classify.The experimental results show that this approach on 2015SNLI dataset gets an accuracy of 0.878.Compared with the best result of 0.889[2]with a difference of 0.011,the neural network model has fewer parameters,and on the RTE7 dataset gets an F-Score of 0.420,the recall is 0.518 which outperform competition optimal value’s recall(0.491).The main contributions of this paper are as follows:1.This paper proposes a method that uses the attention mechanism to align the text softly to solve the diversity problem of the English grammar structure.This method makes the semantic analysis of the text no longer depend on the analysis of the grammatical structure,and partly solve the problem the lexical and syntactic complexities in English textual entailment recognition.2.This paper proposes a method for English textual entailment recognition based on deep neural networks by combining the advantages of various neural network models.Compared with the traditional machine learning method,this method does not rely on the extraction of artificial features anymore,and automatically extracts semantic features via neural network,which is more conducive to recognize textual entailment.3.This paper proposes an ensemble learning method that combines the advantages of traditional machine learning with deep learning.This method fuses the features extracted automatically by neural network with the artificial features in traditional machine learning to improve the diversity of features and then improve the accuracy of textual entailment recognition.