Causal Relation Extraction Of Uyghur Events

As a semantic relationship in the external association of events,causality is both common and important in text,and has a wide application prospect.It reflects the relationship of the succession of events,besides the between the causes and the results.The identification of causality is of great significance for text event extraction and deep semantic understanding.It helps to understand the process of event evolution and the occurrence of events,so as to provide important information for decision-makers to predict the development of later events.In view of the fact that traditional methods can not effectively extract the causal relationship between events in Uyghur,according to the contents and steps of study,we mainly study Uyghur part of speech(POS)tagging and causal relation extraction between events.In Uygur part of speech tagging,based on the causality extraction between Uygur language events,POS tagging set is expanded by combining the existing part of speech tagging set.POS tagging can be regarded as a sequence tagging problem.conditional random field and long short time memory network are the most commonly used sequence tagging model.In the course of practical research,the characteristic function of the conditional random field is very important to the final performance which needs domain experts to elaborate and the final size of the model is large;at the same time the final output vector of the memory network unit can be regarded as a representation of the input data.At last,softmax are generally used to predict the labels,but is has a strong relationship between the direct label processing output data which leads to the effect is bad,especially in the actual sequence labeling task,because the structure of neural network greatly depend on data quality and amount of data will seriously affect the effect of the training model.In order to solve these problems,a hybrid neural network model based on bidirectional long short-term memory network and conditional random field is proposed in this paper.The experiments show that the values of P,R and F reached 90.48%,85.32% and 87.36% respectively.Compared with conditional random fields,the hybrid model is effective for Uyghur POS tagging.In Uyghur event causation extraction task,the traditional way is to extract parts of speech,entity and syntactic information and then bring it into shallow machine learning model.it fails to fully consider the deep semantic information of the event sentence,and the experimental results also show that the traditional method can not effectively extract the causal relationship between events.Therefore,we propose a method based on bidirectional long short-term memory network to extracte causal relation between Uyghur events,which transforms the problem into event pair classification.In order to make full use of the event structure information,11 characteristics of the Uyghur events structure information are extracted based on the study of the events causal relationship and Uyghur language features;besides this paper introduced the word vector as the input of network bidirectional long short-term memory to extracte the deep semantic features of Uyghur sentence events in order to make full use of the deep semantic information event sentences,at the same time in order to accelerate the convergence of the model,we imply batch normalization on the moddel;finally concatenating these two kinds of features as the input of the softmax classifier to extract the Uyghur events.The experiments show that the values of P,R and F reach 89.19%,83.19% and 86.09% respectively,which verifiy the validity of this method in Uyghur event causality extraction.