Research On Time Expression Recognition And Standardization For Chinese Text

In the text of natural language,there is a large amount of time expression,which is an important semantic carrier in the text.Time information can effectively grasp the development of things,so the study of time information has a very important significance for the understanding of natural language.However,there are various forms of time expression,so it is necessary to standardize the time information in order to make the computer process the time information efficiently.Many tasks in natural language processing are dependent on the processing of temporal information,such as question answering system,relation extraction,event extraction,topic tracking and detection and other fields all play an important role.This thesis focuses on the research status and problems of time expressions at home and abroad,put forward in time expression recognition will advance training language model is introduced into the time of information processing,can get a lot of prior knowledge,solve encountered in the past using a loop neural network single word vector and long distance depend on the capacity of poor phenomenon,so introduce BERT language model,solve the distance dependence,and provide polysemous word vector for subsequent model,the experimental results show that the effect is remarkable.Inspired by the fact that time dictionaries are often added to the rule system,this thesis proposes to use the FLAT model and BERT model to introduce the time lexical features into the neural network model,which can also improve the recognition of time expressions.In terms of the standardization of time expressions,this thesis mainly studies the multi-label classification of time expressions encountered in previous research work,and improves it so as to better deal with the standardization of time expressions.To use the language training model of dynamic character vector and the combination of SGM model to solve encountered in the process of standardization of tabbed classification problem,and the time expression recognition and standardization in the model of the contact,the retention time expressions in the sentence semantic information,and the experimental results show that the method for time expression standardization process encountered many labels have better improve classification problem,model is better than other algorithms.Finally,Chinese time expression recognition and standardization system based on Web application is designed according to the proposed algorithm.User input in the Web client contains time expression statements in Chinese,after front passed into the intermediate control layer for data preprocessing,will process after the completion of the model analysis of data from the back-end processing module,according to the corresponding rules will time expression is mapped to the corresponding time axis,and return the results to the Web client.