National Security Microblog Search And Topic Detection Based On Deep Neural Network

Microblog is a broadcasting social networking platform based on the user's attention mechanism to share,exchange and send short and real-time information.With the growing popularity of online social networks and the large number of active users over there,online social networks have accumulated a wealth of user-generated data,including cross-media data such as text,images and video,which contain vast amounts of valuable information.The diversity of Weibo information service and the diversity of users' demands for information acquisition make the search service on Weibo different from the traditional Internet search service.This makes it a hot topic to study the microblog search algorithm which is more efficient and can meet the needs of user information.The main work done in the thesis is as follows:(1)The Weibo text representation method based on recurrent neural network is proposed.The structure of recurrent neural network based on LSTM unit is improved,which makes it possible to obtain a more accurate potential semantic representation of Weibo texts through training process.The experimental results show that the proposed microblogging text modeling algorithm can extract the potential semantic relationships of microblogging texts effectively,and the semantic inference experiments show that the accuracy of semantic inference algorithm is greatly improved compared to word2vec method.(2)The microblog search method based on deep recurrent neural network(NBSM)is proposed.The algorithm considers the text relevance and the social influence factors,and proposes a method to calculate the similarity of Weibo texts based on recurrent neural network language model,and calculation method of social network influence based on the popularity of microblog,and the use of word similarity to expand query words in the microblog search term.Experimental results show that compared to CNNM method,which based on convolutional neural network,the proposed NBSM method based on deep recurrent neural network improves the accuracy of microblog searching results over MAP values and NDCG value.(3)The topic detection method based on deep convolutional neural network(DCNN)is proposed.Based on the microblog text information,the deep convolutional neural network theory is used to establish the microblog topic detection model based on deep convolutional neural network,which can fully exploit the potential semantic features of the Weibo texts.The experiment of Twitter sentiment analysis is designed and implemented.The experimental results show that compared to CNN method,which based on convolutional neural network,the proposed DCNN method greatly improves the accuracy of twitter sentiment analysis.The experiment of Weibo topic detection is designed and implemented.The experimental results show that the compared to CNN method,the proposed DCNN method greatly improves the accuracy of Weibo topic detection.(4)Design and develop a national security related microblog search and topic detection system.The system includes the Weibo text feature extraction module based on deep neural network,the microblog search module based on deep recurrent neural network and the microblog topic detection module based on deep convolution neural network.The system has perfect functions and meets the requirements of Weibo search and topic detection system.The thesis propose the Weibo text representation method based on recurrent neural network,the microblog search method based on deep recurrent neural network and The topic detection method based on deep convolutional neural network.Design and develop the national security related microblog search and topic detection system.The system has perfect functions.