Study On Chinese Temporal Expression Recognition And Normalization

Individual perception about time plays a role in understanding the development of things.Specifically,time information extracted from text contributes to knowing the context which has also proved to be of significant value in information extraction,question and answering system,text summarisation,topic tracking and detection in the domain of Natural Language Precessing(NLP),whereas time expression recognition and normalisation is the principal foundation.This paper studies and analyzes the overseas and domestic research status,methods and existing systems,respectively,in time expression recognition and normalisation;furthermore,a method based on neural network to solve the problem for Chinese expression is proposed.In regard to time expression recognition,BiLSTM-CRF model adopted in this paper autonomously abstracts features in stead of manual building as well as selection.Generally,BiLSTM-CRF model has reached a level comparable to traditional CRF model,and even has a higher F1 score up to 84.25% according to TempEval-2010.Moreover,this paper aims at reducing the complexity of time expression normalization by categorization based on word vector and neural network multi-labelling method which is superior to traditional multi-labelling method.In conclusion,this thesis applies two neural networks to address the problem in Chinese time expression recognition and normalization independent of manual feature building and selection in traditional method.The result indicates that both networks have a good performance.