New Event Detection Based On Mongolian News Corpus

As the main language used by the main ethnic groups in the Inner Mongolia Autonomous Region,Mongolian has been widely applied in various fields.Users are distributed in countries such as China,Mongolia,and Russia,which have a certain influence in the world.With the gradual increase of Mongolian-related network resources,the demand for detection and tracking of Mongolian information is increasing.However,So far the new event detection task of Mongolian texts is still blank at home and abroad.Therefore,it is urgent to conduct research on the new event detection task in Mongolia.This paper proposes a new event detection method based on deep learning,aiming at the detection of news events in Mongolian news corpus.First of all,this paper applies the traditional vector space model(VSM)to new event detection based on Mongolian news corpus tasks.This method is mainly used to model document content and used as a baseline method.Secondly,the Named Entity Recognition(NER)is integrated into the traditional vector space model to further extract the location and organization name of the document.Compared with the traditional VSM model,this method has a lower false detection rate.Then,in order to extract multi-dimensional event elements which including subject,object,time and place,Bidirectional Recurrent Neural Networks(BiRNN)is used to extract event elements.The experimental results show that compared with the VSM-NER method,the false detection rate and false detection rate are significantly reduced.In order to further improve the accuracy of the detection of news events in Mongolian news corpus,the method of combining bidirectional recurrent neural network and conditional random field(CRF)is applied to the detection of new events,as a result,the detection performance is further improved.