Design And Implementation Of Chunk Parsing System Based On Deep Learning

Syntactic parsing,as a key and difficult problem in natural language processing(NLP),has always been the focus.As an important part of shallow syntactic analysis,chunk parsing reduces the task difficulty of full syntactic parsing.The chunk parsing system is mainly to identify and classify chunk information in a sentence.Chunk information can be used as an intermediate product in many areas of natural language processing,such as information retrieval,information extraction,text classification,speech recognition and so on.With the popularization of artificial intelligence,deep learning technology has achieved fruitful results in many fields.Of course,deep learning techniques are also very effective in natural language processing.A series of network models based on recurrent neural networks(RNN)can handle sequence data of arbitrary length well.A lot of practices have proven that deep learning algorithms are superior to statistical machine learning algorithms in processing many tasks in the field of natural language processing.In this paper,the deep learning algorithm is applied to the chunk parsing task to complete the design and implementation of chunk parsing system based on deep learning.This paper introduces the background and basics of chunk parsing and related deep learning algorithms.According to the actual application situation,the analysis of business process and functions of the system,the system is divided into two subsystems.Model training subsystem and chunk recognition subsystem.There are multiple modules in each subsystem to achieve different functions.The chunk parsing system mainly includes a data pre-processing module,a word embedding module,a model training module and a chunk recognition module.There are clear boundaries between modules.The functions of modules are clearly defined.The modules can be connected together and form the whole parsing system.The implementation of main modules and main functions are described in detail.Then,according to the functional requirements of the system,functional tests are performed on main functions to ensure the safety and stability of the system.This paper compares deep learning models with traditional machine learning models in processing chunk recognition tasks.On the same training sets and validation sets,the maximum entropy model and support vector machine model achieve precision rate,recall rate and F1 of roughly 96% for different chunks.The accuracy rate of the LSTM model was 97.8%,the recall rate was 97.7% and the F1 was 97.8%.It can be seen that the neural network model based on deep learning has completed the chunk parsing task well,and has obvious advantages in the recognition effect of each chunk.