The Research Of Event Relation Extraction Based On Deep Learning

The event relationship reflects a semantic relationship between events,which is both common and very important in text representation.Most of the existing research work on the relationship between events is concentrated on Chinese and English.For this problem,this paper will study the Uyghur event accompany relationship and the temporal relationship.main tasks as follows:(1)In terms of the event accompany relationship,the occurrence of two events is very tight,and one event usually appears with the appearance of another event.For the Uyghur event accompany relationship,a deep Belief Network(DBN)recognition method is proposed.12 event structure features based on Uyghur language features and accompany relationship are extracted,and the word vector information of event trigger words is combined to obtain the semantic similarity of the two events triggers the word.The event structure feature and the semantic similarity are represented as final events.This method is used to identify the accompany relationship of Uyghur events,and the F value is 82.48%.In order to avoid the constraints of complex feature engineering,this paper proposes a long-term short-term memory(LSTM)recognition method based on the theme,which only uses the word vector representation of the event sentence as the model input.It can avoid complicated feature engineering,and it also has a great improvement in recognition effect.The F value reaches 96.72%,which indicates the effectiveness of the method in Uyghur event accompany relationship identification.(2)Aiming at the problem of Uyghur event temporal relationship identification,a Bi-directional long short term memory(Bi-LSTM)based on attention mechanism is proposed to identify the temporal relationship of Uyghur language events.Firstly,combining the characteristics of Uyghur event temporal relationship,13 manual features are extracted,which are called internal structural features of the event.The event sentence vector representation is used as the input of Bi-LSTM to obtain an event representation containing contextual semantic information.The event attention mechanism is established based on the event trigger word to get the event semantic feature of the event sentence is obtained.The internal structural features of the event and the semantic features of the event are combined to be the input of the Softmax layer to conduct the temporal relationship task.The experimental results show that themethod can obtain the corresponding event semantic features while obtaining the implicit semantic information of the event sentence.After the internal structural features of the event are merged,the F value reaches 88.03%,which proves the effectiveness of the method in the Uighur event temporal relationship identification task.