Study On Social Network National Security Threats Source Search And Tracking

Recently,social networks develop rapidly and become more and more important as well as popular in daily life.However,not like traditional media,everyone on social networks is able to release and broadcast information,causing the difficulty in judging the authenticity of information.National security and social stability are facing an unprecedented challenge under this circumstance.There are some suspicious forces that spread rumors and incite riots via Internet at home and abroad.In the research of information dissemination on social networks,it is very meaningful to study search sources of information and tracking users.The main work of this thesis is done as follows:(1)We gather a big amount of data from social network by using crowd sense technologies.This part lays the foundation for searching and tracking.We can get social features,such as publishing time,the number of forward and the number of comment as well as user features from data.Using open APIs,we are able to crawl the micro-blogs and users' attributes within a period from Sina Weibo.We filter and classify the raw data according to keywords related to national security and store the features in a designed database.(2)We have proposed algorithms of opinion leaders mining combined with hierarchical sentiment analysis.We use information flow tree to calculate users' influence score.This information flow tree is a compact structure like Frequent Pattern Tree.The users' threat score is calculated through the hierarchical emotional analysis.After aggregating users' influence score and threat score,we can get a list of opinion leaders who may threaten national security.We optimize the list with influence maximized models to acquire a more accurate result.Experiments show that our algorithm increases nearly 21%in MAP.After the analysis of the maximum influence propagation model,it improves about 5 times in MAP.NDCG is greatly improved as well.The experiment verifies the effectiveness of our methods on mining opinion leaders related national security.(3)We have proposed an algorithm of source searching and topic word identification based on words' location.The topic words and threatening users can be acquired by matching dubious words according to their locations in sentences and the related dictionary.Combined with the network of information dissemination,we are able to derive a set of possible sources of threats and focus on monitoring them.We use the improved SIS model BFS to study the change of users' status and analyze users'behavior while diffusing in social networks.The experiments show that our algorithm increases by 54%in the average MAP.The recall and accuracy are improved by 4%and 25%respectively,which demonstrate the utility of our algorithm on finding sources of threat.(4)We have designed and implemented a search and track system related to threaten sources of national security by combining the above researches.All the results and analysis under different algorithms are displayed in the system.The system is composed of three parts which are data acquisition module,opinion leaders mining module and source searching and tracking module.After testing,we proved the usability and stability of system with functionality and friendly interaction.