| With the rapid development of Internet technology,the network has become the most direct way for people to understand the news of domestic and abroad.Especially with the risen of domestic application,such as TouTiao,NetEase News and so on,more and more people choose to use online news to understand current events.At the same time,due to the virtual nature of the network,some outlaws use the network to spread false information,and a slight carelessness will cause social panic and instability.Therefore,for the network security supervision work,The importance of the detection and prediction of emergencies is self-evident.The paper proposes an emergency detection and prediction method based on improved Autoencoder.When convert news to vector,considering vector dimension of document is high and sparse,the paper proposes a K-Competitive Autoencoder with R-Threshold(KAER).Use neurons in hidden layer to represent reduced-dimensional text vector.For another,with problem of topical heat value of event in emergency prediction model based on aging theory is affected by timeslot,the paper proposes an improved energy attenuation scheme,which makes it more suitable for actual number of news texts to change with time.The main work and innovations of the paper are as follows:(1)By analyzing merit and demerits of current text vector representation method,combined with the characteristics of news document,the paper uses a dictionary-based space vector representation model.Based on the high dimensionality and sparsity of vector of news,KAER is used to reduce the dimension of vector,and adopts the improved Sinle-Pass method for the clustering of news vector after dimension reduction.(2)By analyzing the existing emergency detection methods,the paper proposes an emergency prediction model based on aging theory,and constructs a network news hot event and emergency prediction model by improving the energy attenuation calculation method of the topic and combining the improved growth rate-based prediction method.(3)The thesis is based on real and reliable news data,the feasibility of the proposed text vector dimension reduction method are verified by comparative experiments,and evaluates the reliability of the improved subject heat value calculation method. |
