Social Network Cross-media National Security Semantic Learning And Microblog Topic Search

With the popularity of social network,more and more users transmit information through social platforms,which makes a large number of data spread quickly and widely on social network.The types of data in the social networks are very complicated,like text data,image and video data.Therefore,study how to extract the semantic information from various types of data and mapping the features of different data types into a unified semantic space will have great significance.At the same time,users'demand for information retrieval on social platforms is increasing.Facing the problem that the number of words on social platforms like micro-blog,the research of using query expansion methods to promote information searching has become a hot topic.The main work accomplished in this thesis is as follows:(1)Convolutional neural network is used to extract the deep features of data from social networks.By improving the original AlexNet network structure,a new structure named AlexNet-Social is proposed for feature extraction of social network data.AlexNet-Social can extract the deep semantic information of data more effectively and it can reduce the work of parameters calculation.Experiments on social network image data sets show that the classification accuracy of the proposed algorithm is improved by about 5.6%and the training efficiency of the model is improved by about 36%.(2)A cross-modal semantic model based on AlexNet-Social for national security events in social networks is proposed.The model uses deep neural network to extract the features of text and image data respectively,and attention mechanism is used to get the unified vector expression of the two different types of data.Using social network text and image data sets to test the proposed model,experiment results show that the CSMBA model has the best performance in the accuracy,recall value and F value evaluation indicators of related event recognition tasks.(3)Combining social characteristics and time factors,a Weibo searching algorithm WSAST is proposed.In terms of query word expansion,the combination of semantic similarity and temporal similarity is proposed to expand the query words,which not only utilizes the semantic information in the microblog text,but also considers the temporal distribution of words.In terms of rearrangement of search results,textual word frequency and microblogging hot point rearrangement are used to make full use of the social characteristics of Weibo users to optimize search results.Experiments show that the WSAST algorithm has the best performance in the search accuracy index compared with other comparison algorithms.The WSAST algorithm effectively improves the performance of Weibo search and better meets the user's search needs.(4)A Weibo security event identification and topic searching system is designed and developed.According to the demand analysis,the system includes the five functional modules:Weibo feature extraction module,the Weibo searching module and the Weibo topic detection module.The system has perfect functional performance and meets the requirements of Weibo searching and topic detection.