Research On Natural Language Text Reading Comprehension

With the rapid development of artificial intelligence and big data,teaching computers to understand human natural language has become the most challenging complete goal of AI.Let the machine learn to read the text like a human.By understanding the internal meaning of the text,it can give the correct response and feedback according to the problems and needs.This task process is called the machine reading comprehension task.Machine reading comprehension,as an important research area in natural language processing tasks,has become a focus of attention in academia and industry.With the development of deep learning technology and the continuous release of large data sets,the difficulty and challenge of machine reading comprehension tasks have gradually increased.Research on machine reading comprehension also depends on other technologies related to natural language processing,which promotes the development of the entire natural language processing field and then affects the progress of artificial intelligence.This makes machine reading comprehension very important for research.This paper aims at the Chinese machine reading comprehension dataset Du Reader.By analyzing the characteristics of the data set,this paper proposes machine reading comprehension methods based on feature engineering and deep learning respectively.For data with obvious characteristics,a method based on feature engineering was used to design a text semantic analysis model.Methods such as pronoun analysis,marked basic noun phrases,and syntactic analysis was used to enrich the semantic information of documents and sentences,and synonyms were added using the external knowledge base CSC.The set of words to expand the question set is used to mine semantic relevance and expand the scope of the answer to obtain an exact match between the question and the correct answer sentence.In addition,in order to further obtain accurate answers,this paper designs text semantic reasoning technology based on the text semantic analysis model,defines seven frame semantic relationships to represent the connections between scenes,and obtains some potential semantic information,and constructs more than 30 features are used to identify the core target words in the sentence.Narrow the scope of answers expanded by the expanded vocabulary through scenes activated by the core target words.Combining core target words between frame relationships to improve the accuracy of answer acquisition.For features whose features are not obvious or where feature data cannot be obtained directly,this paper uses a method based on deep learning.This paper draws on the ideas of the Bi DAF model and proposes a self-attention-based SAR_Bi DAF machine reading comprehension model.On the basis of the original two-way attention,a word attention mechanism and a selfattention mechanism are added,and a multi-step reasoning mechanism is introduced in the semantic matching process to accumulate the focused information to achieve the optimal.The answer generation method is generated in two modes: generation mode and extraction mode,further improving the accuracy of answer generation.Finally,the method based on feature engineering and the method based on deep learning in this paper are experimentally verified on the Du Reader dataset.The experiment is mainly divided into two processes: feature engineering experiment and deep learning experiment.For the evaluation of experimental results,ROUGE-L and BLEU-4 are selected as the evaluation indicators.By comparing experiments with the baseline model,the experimental results show that the model proposed in this paper has better performance and generality in machine reading comprehension tasks.