Classification Of Temporal Relationships Between Chinese And English Events Based On Deep Learning

The classification of event temporal relationship is an important task of natural language understanding,aiming at mining the relationship of events in the text in the time dimension,exploring the sequence of events,and classifying them.Events are often described as changes in state,with attributes such as trigger word,subject,object and type.The evolution from one event to another has internal laws and relations,and the temporal relationship of events is the connection in the time dimension.The study of the temporal relationship of events plays an important role in understanding the law and logic of the development of events,and also plays a great role in the fields of information extraction,question and answer system,and decision-making system.For example,the logic map integrated with the temporal relationship has a more rigorous structure,clearer relationship context,more accurate and clear presentation of the development trend and evolution law of events,and can assist the question answering system or decision-making system to give better answers.In the field of medicine,the automatic construction of time schedule and the formulation of diagnosis and treatment plan for patients can reduce the workload of medical staff to a certain extent and focus on more important matters.In the early stage,the classification of event sequence relations mainly adopted rule-based and machine-learning methods.The former overly depended on the quality of data annotation,requiring the annotation personnel to have relatively rich domain knowledge and receive certain linguistic training.Although traditional machine learning methods reduce the requirements for data annotation personnel,they perform generally in text understanding tasks.The deep learning method can learn advanced features from the data center and deeply mine potential information,greatly weakening the dependence on the quality of data annotation and reducing the labor cost.(1)A Timestamp-Enhanced Event Temporal Relationship Classification Method in Hyperbolic SpaceIn order to capture the rich hierarchical structure between event texts and resist the interference of external knowledge base on the local corpus,this paper proposes a timestamp enhanced classification method of hyperbolic space event temporal relations,TEETR.After mapping from Euclidean space to hyperbolic space,the relative time stamps between event pairs are trained to resist external noise,and then the event temporal relations are classified by combining the characteristics of external knowledge base.The experiment shows that the recall rate and accuracy rate of the method in this paper are 1%higher than those of the baseline task in the MATRES dataset,and 2%-3%higher than those of the baseline task in the TCR dataset.(2)A method of Chinese event sequence classification based on graph convolution neural networkIn the task of event temporal relationship,the Chinese dataset is relatively scarce compared with the English dataset,resulting in the relatively lagging research of event temporal relationship classification of Chinese text.In this paper,the classification of event temporal relations is explored on the Chinese emergency corpus CEC.The traditional deep learning method is good at processing data in European space.There may be such multi-dimensional features as attribution and dependency between events,which can better reflect the meaning of sentences in non-European space.For this reason,this paper proposes a Chinese event temporal relationship classification method based on graph convolution neural network,which constructs one-dimensional text statements into multidimensional graph structure data,and trains in the convolution neural network layer to complete the classification work.Experiments on CEC data sets show that the performance of this method is better than other baseline tasks,and better experimental results are achieved.