| In recent years,Natural Language Processing(NLP)has developed rapidly in the field of artificial intelligence.NLP has become an important direction in the field of computer science and the development of artificial intelligence industry.The main objective of NLP is to make human and computer freely communicate with natural language.The machine learning is commonly used for NLP,especially statistical machine learning.Recently,deep learning has been the mainstream of NLP.The researchs about the Natural Language Inference(NLI)and the Cloze-style Queries are the most important foundation of deep learning applications in NLP.This paper analyzes the Natural Language Inference in NLP.Then we proposed a deep learning model based on semantic discarding and semantic aggregation.In this work,we first make two important observations regarding NLI:(1)the existence of extra/interfering semantics and its negative impact on the correctness of final inference;and(2)the unbalanced importance of local inference results and the need to combine all local results for aggregation.Motivated by these two observations,we have designed a Semantic Drop Network(SDN)to discard interference information autonomously.Also,We designed a Semantic Fusion Alignment(SFA)method to fuse the results of local inferences more effectively.Then we designed SDF-NN,a new NLI model.Through this model,the accuracy of the relation judgment is improved.For the reading comprehension of cloze,we proposed a deep learning model based on memory modification.We first analyzes previous attention based models and reasoning based models.In these studies,it has been found that the layers of network structure is gradually deepened,so that the model can be capable of reasoning and thinking like human beings.Although the model can extract more information features through the deepening of network structure layers,deep networks can also have negative effects.The deeper the network layer,the weaker the memory of the original information。With the deepening of the network layer,the focus of attention may be deviated,and thus output wrong answers.To solve this problem,a memory modification mechanism is proposed in this paper.In the iterative inference process of network,the global memory information is added.And we proposed mGA Reader,a deep learning model based on memory modification for cloze reading comprehension.In the process of network iterative learning,each layer of encoding combines the original document information for correcting attention to prevent a large range of attention deviation,and finally output the answer.In the problem of reading comprehension,the model improves the accuracy of outputting the correct answer,and verifies the validity of the memory modification mechanism. |
