Research On Deep Learning And Models Ensemble Based Argument Reasoning

With the rapid development of artificial intelligence and Natural Language Processing(NLP)research in recent years,people have put forward new requirement on the degree to which the computer understands the natural language.The Argument Reasoning Task is one of the basic tasks of the natural language understanding,together with information extraction,question answering,machine understanding,text analysis and other fields of Natural Language Processing.Therefore,the Argument Reasoning Task is concerned by the research institutions and researchers of the world.Argument reasoning is an important part of the natural language understanding.In this thesis,we introduce a natural language argument reasoning task and related data set,given a Reason and a Claim,from the two Warrant to select a correct implicit Warrant,make the Reason to support the Claim,to form a reasonable syllogism structure(Reason,Warrant,Claim).Based on deep learning methods,we use the pre-training vectors,Word2 vec and Glove,construct word vectors for Reason,Warrant and Claim,use word vectors representing the syntax and semantics of the text information.A series of end-toend reasoning systems are constructed by using a series of deep learning models such as Convolution Neural Network(CNN),Long Short-Term Memory model(LSTM),Bidirectional Long Short-Term Memory(BiLSTM)and Attention Mechanism,systems reasoning ability are compared through experiments.Finally,model integration is carried out through the ensemble method to improve the accuracy of the single system and improve the generalization ability.The accuracy of 67.09% on the validation set and the accuracy of 56.21% on the test set are achieved through the deep learning model,and the accuracy of 68.67% and the accuracy of 56.63% on the test set are achieved by the ensemble method.The experiment proves that the method based on deep learning model and ensemble method can better solve the argument reasoning task of natural language argument.