Semantic Reasoning Based On Deep Learning

Argument reasoning is a foundation technology in natural language processing,which can help us to better understand the content of the article and the semantic relationship of the context.Argument reasoning is generally divided into two parts: the reasoning and the warrant.Reasoning is the context that support the warrant,which used to prove the reason of the arguments.The argument is an understanding of the reasoning which is discussed by essay,and it is an accurate summary of the article.Argument reasoning task's definition is judging whether the reasoning and the warrant is entailed.Improving the accuracy of argument reasoning can increase the ability of understanding natural language in the field of artificial intelligence.In the early stage of natural language processing,people usually use dictionary,syntax tree or logical expression to obtain a shallow inference relationship between sentences.With the application of neural networks in natural language processing and the rapid development of attention mechanism and transfer learning,researchers began to focus on mining deep semantic inference relationships between sentences.And researchers hope to speed up and optimize the learning efficiency of the model by transfer knowledge between similar tasks.This paper mainly studies the argument reasoning technology from the following three directions:1.Argument Reasoning Technology Based on Long Short Term Memory Network.We use the long short term memory networks to code the reason and the warrant.Then,we use bilinear or deep neural network to judge the inference relationship between sentences.2.Argument reasoning technology based on long short term network and attention mechanism.First,we use long short term memory networks to code the reason and the warrant.Then we use attention mechanisms to focus on key words of the reason and the warrant,which make models have stronger semantic representation capabilities.Finally,we use the semantic vector which encoding by long short term network and attention mechanism to judge the reasoning relationship by bilinear or deep neural network,and predict the inference relationship between premise and inference sentences.3.Argument Reasoning Technology Based on Transfer Learning.Let the reasoning model learns rich semantic vector representations and semantic inference representations from large-scale inference datasets using transfer learning technology.And then transfer the knowledge from source task to target tasks,which makes the reasoning model of the target task can be quickly fitted.This paper also use the ensemble model which combine the traditional features and deep learning and mining semantic inference relationships between texts from a variety of perspectives.