National Security Information Search In Online Social Networks Based On Cross-media Semantic Features

With the growing number of social network users,more and more data are generated in the social network every day.However,the data in the social network often have problems of noise,diversity and semantic sparsity.Therefore,it should spend a certain amount of effort to make good use of the valuable social network information.This thesis aims at the common issue of social network data mentioned above,mainly studies in the following four research points,acquisition of social network cross-media spatial-temporal data,semantic modeling of cross-media spatial-temporal information,cross-media search in common semantic space of social network data,and designing the national security information search system of online social networks based on cross-media semantic features.The specific work is as follows:(1)In the field of acquiring social network cross-media spatial-temporal data,a method of acquiring spatial-temporal data of social network national security information is proposed.In order to solve the problem of noise and diversity which is widely existing in the social network.We establish the national security keyword database which is used to acquire the national security information in the social network.by using the security keyword database,the relevant social network cross-modal information is obtained,including the micro-blog text,image,and other spatial-temporal information.After acquiring the original data,the noise information widely existing in social network data is filtered.(2)In the field of social network cross-media data semantic modeling,a text semantic modeling algorithm based on spatial-temporal topic embedding(STTE)and a spatial-temporal cross-media semantic correlations modeling algorithm(STECM)are proposed.To model the semantic information of the social network short text data,the STTE algorithm exploits the global spatial-temporal information and local context information,models the topics of the spatial-temporal region,and obtains the precise social network spatial-temporal topic feature.Compared with the traditional probabilistic topic model,the classification accuracy of spatial-temporal topic embedding is improved by 12.7%.Compared with the traditional word embedding method,the performance of spatial-temporal topic embedding is improved by 9.2%.As for the image data in the social network,the convolutional neural network is employed to mine its deep semantic information,the semantic mapping relation between cross-media data is established by using the cross-modal correlation mapping function.The classification performance of the cross-modal features is improved by 6.9%for the national security data of social networks.(3)In the field of social network cross-media data semantic search,a cross-media semantic mapping function based on the deep random walk(DWM)is proposed to solve the semantic gap between data of different modal.By mining deep semantic relation between social network cross-modal data,DWM improves the semantic mapping accuracy.Furthermore,a deep hashing semantic cross-media search algorithm(DHCS)is proposed.By employing the semantic extension and deep quantization hashing,we can achieve a better performance of cross-media search on social network data,the MAP index is improved by 13.1%,and the P-R curve is improved by 21.2%.(4)An online social network national security information search system based on cross-media semantic features is constructed.It includes social network national security data acquisition module,social network national security semantic modeling module and social network cross-media semantic search module,the online social network national security information search is achieved.