| Social media revolutionizes the information communication,and enables every user to become a publisher rather than only a receiver.Through social media,users can enjoy instant messaging and information,and share news and views anywhere and anytime.Along with its development and population,social media seems to dominate the information publication,thereby obliging the mainstream media to follow,which has placed the social management in a passive position with potential risks.Therefore,to keep society stability and personal information safety,security management of socail media has become an important subject for domestic scholars.The hot topic dynamics in social media are closely related with the public opinion.To facilitate social information management and follow public opinion dynamics,the topics in social media need to be supervised and analyzed timely and effieciently.Over the past years,the research about social network paid main attention on the quality of topics,which usualy based on complicated calculations.However,the topics in social media are decentralized and switching rapidly,making it difficult to catch the instant semantics of social media topics through traditional strategies.Therefore,how to filter the useful knowledge efficiently to meet the real-time requirements without sacrificing topic quality from the distributed network nodes is becoming a challenging task.To deal with these challenges and target the source of hazarded factor quickly,this dissertation focused on the problems of burst topic detection and data source location based on text data stream in the social network.The results in this study consists of five aspects as following:(1)To solve the problems of spatial-temporal dependence and semantic sparsity in social network,we performed the expression and semantic analysis of spatial-temporal data in social network.Based on the analysis of the correlation between online topics and offline activities,as well as the factors that affect modeling of topics in social network,we proposed a background feature fusion(BFF)method.Specifically,the feature words are respectively expressed as background features to represent the spatial-temporal information.Then,a tensor is obtained by BFF under the constraint of data structure,where each face of the tensor represents the distribution of feature words in spatial-temporal and structure.Fusion operation allows mitigation of semantic sparsity problem.Online topics are connected with offline activities due to the spatial-temporal association.Based on BFF,social topics are identified by feature words clustering.The BFF-based algorithm is convenient for topics identification with various granularity through clustering,which is the basis of the following burst topic detection and data source search.The experiments show the preferable effects of BFF-based method in the validity and the reliability.(2)To solve the problem that how to detect potential hot topics in social network at early stage,the real-time detection of burst topic is studied.Through the investigation of the evolution low of text data stream affected by users' behaviors as well as the influence by uncertain factors,we proposed a real-time monitoring method of social data stream based on denoisinging.Burst factors are identified on the basis of data prediction.In stread of passively waiting for an emergency topic to sause our attention,we try to detect the potential burst topic in its budding stage from the data stream.This method tracks the temporal features of keywords throung real-time monitoring of data stream,and predicts the possible development trend of keywords in the future through the historical records of data.Thereby,the abnonnal keywords are identified as potential burst factors to trigger burst topic discovery.The effectiveness and robustness of the proposed method are also tested.Experimental results show that the proposed method can detect the burst factor 1-time unit ahead of the comparison methods from the breaking topics,i.e.,10-100 seconds in the experiment,and advanced from the developing topics,i.e.,more than 10 time unit in the experiment.Therefore,the proposed method can satisfy the in-time requirement of topic detection.(3)To satisfy the real-time processing restriction,fast discovery of spatial-temporal topic in social network is studied.This dissertation analyzed the spatial-temporal characteristics of social topics influenced by historical data,and proposed a algorithm of Fast Discovery of Burst Spatial-temporal Topic(FDBST)based on the trend prediction of data stream evlution.FDBST aligns data prediction with characteristic calculation to detect burst term and integrates region topic detection with global topic detection to find spatial-temporal burst topic.Specifically,the online topics are constructed into strong connection graphs and the global topic is obtained by local topics fusion in which the strong connection graph including the burst factor is the burst topic.Instead of generating a complete topics' model for the dataset,FDBST constructs a separate model for the burst one.Therefore,it keeps the interference caused by the noise data off,and has a significant improvement in computing efficiency.Experimental results show that preferable effects are procured,and FDBST outperforms state-of-the-art approaches in terms of effectiveness,i.e.,FDBST controls the delay within a 0.1 seconds level while preserving the topic quality.Therefore,it can meet the real-time requirements.(4)To address these curity problems brought by online topic evolution,data source search of hot topic is investigated.This dissertation analyzed the structural change of online topics caused by user behavior in the process of evolution,and proposed a graph-based method for Discovery of Hot Topic Sources,namely DHTS.DHTS searches the related data on the time line in reverse according to the structure of the specified content,i.e.,the security topic.Specifically,the public topic is modeled as complete graph,then the relevant local topics are found by constructing strongly connected graph.Based on that,DHTS keeps track of public topic according to structural variations of graphs.Finally,the data source is decided together by global topic and local topics.DHTS does not pursue content similarity but relevance,so the results are not necessarily similar to the search content but closely related.Experimental results demonstrate the effectiveness of the proposed method.They show that compared to the baseline,DHTS has better performance in Novelty and Time,and worse performance in Coherence and Correlation.The continuous distribution of correlation values on the timeline proves the accuracy of data source.Therefore,the proposed method is suitable for content search in different stages of the same topic.(5)An online social network search system based on spatial-temporal and behavioral characteristics is designed and implemented on the basis of spatial-temporal topic discovery algorithm,FDBST algorithem and DHTS algorthem.This system consists of threemodules,which are the discovery module of spatial-temporal topic,the real-time detectiommodule of burst topic and the data search module of security topic source,to realize the discovery of spatial-temporal topics,detection of burst online topics and search of security topic data sources.The discovery module of spatial-temporal topic returns the spatial-temporal topic related to the input content in a specific location according to the users' need.The discovery module of spatial-temporal topic shows users the temporal characteristics of spatial-temporal data stream and the structure information abnormal data.The data search module of security topic source shows users the reverse change process of relevant information on the timeline according to the input content,and returns the final data results.In summary,this dissertation provides an online social network search method based on spatial-temporal characteristics and user behavior characteristics for the security management of social networks,which can be used for the instant detection of online hot topics and rapid search of security topic sources. |
PDF Full Text Request