Chinese Syntactic Analysis Research Based On Deep Learning

With the rapid development of artificial intelligence,natural language processing has become more important as a key technology for the exchange of human beings and computers.Natural language processing research includes lexical analysis,syntax analysis,semantic analysis and so on.According to the different levels of analysis,Natural language processing can be divided into two kinds of shallow analysis and deep analysis.Shallow analysis techniques are mainly applied to the vocabulary level,and deep analysis technique is to deal with the grammar,semantics and pragmatics of the whole text.Given the grammatical system,syntactic analysis can infer the grammatical structure of the sentence automatically and analyze the relationship between the grammatical units in the sentence,which belongs to the deep analysis technique.The lexical analysis technology which is at the bottom of syntactic analysis is basically mature.The deep semantic analysis technology is based on syntactic analysis.Syntactic analysis plays a crucial role in the process of natural language processing and is at a crucial position in natural language processing.Although the traditional syntactic analysis model has higher performance,it often uses a large number of manually selected features and their combination,which requires a lot of labor costs and serious dependence on the experience of the model implementer.The advantage of the model based on deep learning is that it can automatically extract features for modeling,makes full use of the neural network's own feature learning ability,and avoids a lot of manual participation.The syntax analysis model based on deep learning has much higher performance in natural language processing than the traditional syntax analysis model.The introduction of deep learning method into syntactic analysis has now become a new research hotspot.Based on the analysis of Chinese syntactic features,this paper introduces a deep learning model to Chinese,syntactically analyze,designs a more effective neural network structure,and uses a more accurate feature extraction and calculation method.The main work is as follows:(A)Proposed a graph syntax analysis model.The model has a more generalized network structure.The bi-directional long-term memory neural network model is used to generate the vector representation of the observation probability in the hidden Markov model,and then the constraint conditions are added to ensure that the analysis results are more in line with the language's own logic.Finally,Hidden Markov Model gets parse tree.The neural network model defined in this paper can generate the observed probability vector effectively,and the reasonable filtering procedure can avoid the syntactic analysis result of the inconsistent language logic.The Hidden Markov Model can effectively carry out the classification work.Experiments on short sentences and long sentences are carried out on the Chinese Treebank Library.The experimental results show that the syntactic analysis model proposed in this chapter can significantly improve the performance of the syntactic analysis.(B)Proposed a progressive learning method based on the syntax analysis method.Due to the rarefaction of data and the limitation of space and time in terms of computation,the usual parsing methods use either unit grammar or binary grammar to replace global context information.This paper presents a new method for calculating probability: Method for calculating the probability of trees.Compared with the traditional chain probability calculation method,it can better balance the role of various feature items(especially sparse feature items)in the overall probability value calculation.According to this probability calculation method,Grammar models or binary grammar model of its ability to obtain global context information stronger.Papers apply it to the classification of syntactic tags.Aiming at the hierarchical relationship between syntactic structure and syntactic tag,we first analyze the syntactic structure,and then classify syntactic labels according to the corresponding structure,then introduce a hierarchical model.Most traditional syntax models based on neural networks use a common method of synthetically modeling syntactic and syntactic labels.However,syntactic analysis itself is inherently hierarchical,and tree-based probabilistic computations can use hierarchical structure of parent-child hierarchies of syntactic structures,can give a great deal of help to the classification of syntactic tags.Experiments show that for the classification of syntactic labels,the gradual method proposed in this paper achieves good performance in the classification accuracy of syntactic labels.At the same time,the method can also be applied to other applications that can be separated into multiple steps for natural language processing to extract global features and perform probability calculations.In summary,this dissertation explores the improvement of the performance of Chinese syntactic analysis,designs a highly efficient and useful model for deep learning syntactic analysis,applies a new method of calculating tree densities to the classification of syntactic labels,and greatly improves Chinese accuracy of syntactic label classification on Dependency Syntax Analysis.The work of this paper has made some preliminary achievements.It is the next step to further optimize the structure of the deep learning model and improve the quality of the corpus data set.