Research And Application Of End-to-end Chinese Text Generation

Natural language generation is dedicated to the automatic generation of high quality natural language based on different types of input information.In recent years,the research of natural language generation has attracted more and more researchers' interest,and many natural language production systems have emerged,including weather forecast generation system and robot football real-time commentary generation system.At present,the research of natural language generation is still facing great challenges,and the quality of the generated results cannot fully meet the needs of practical applications.In order to solve the above problems,this paper focuses on end-to-end natural language generation methods based on deep learning.The main contents of this paper are as follows:(1)A dataset for Chinese text generation task is constructed to solve the problem of lack of research resources for Automatic Chinese text generation;(2)A Chinese language generation model based on neural network is constructed.In the same experimental environment,a comparative experiment is conducted with a variety of evaluation indicators.(3)Based on the Chinese generation model constructed in this paper,the application system of Chinese text generation is developed.The main innovations of this paper are as follows:(1)Data set generation for natural language is the premise and foundation of research on natural language generation.Currently,Chinese databases for natural language generation are very scarce.In order to solve this problem,this paper proposes a data set construction method based on English data sets,which combines machine translation and manual work,and uses this method to construct data sets for Chinese text generation.In order to ensure the quality of the dataset,manual evaluation was carried out,and further processing and standardized operation were carried out on the basis of the evaluation results.(2)Deep learning is applied to Chinese text generation task.First,a Chinese generation model based on recurrent neural network is constructed,and then a generation model based on LONG and short-term memory network and its variant GRU is constructed respectively.Based on the data set constructed in this paper,through several experiments,it is found that the generation quality of the generation model based on GRU is better than other models.(3)This paper designs,develops,implements and tests a Chinese text generation system based on GRU under B/S architecture.