Social Network National Security Sudden Topic Detection?Mining And Discovery Of Evolution Laws

With the increasing popularity of Internet technologies,more and more users have begun to contact social networks and have developed a habit of using social networks.Many daily topics that people care about will spread through social media as quickly as possible.In particular,national security topics are closely related to people's lives.Therefore,when bursty topics appear,a lot of relevant public opinion information will be spread on social networking sites represented by Weibo.In order to be able to quickly and accurately detect related bursty topics,this thesis implements a real-time collection of massive data and a system for the detection,mining,and evolution of bursty topics,and uses natural language processing-related theories to efficiently detect bursty topics.Based on the detected bursty topics,topic mining and evolution rule discovery are realized.The main tasks completed in this thesis are as follows:(1)A method for collecting social security national security information and a neural network-based deep feature extraction algorithm are proposed.Aiming at the problems of semantic sparseness and ambiguity of text data in social networks,a short text augmentation algorithm(UCSE)based on text similarity is proposed.Based on the UCSE algorithm to expand the text,the bidirectional long-short-term memory network is used to further extract the depth features of the text.(2)This thesis proposes a bursty topic detection algorithm(BTDF)based on the identification and filtering of social network bursty features.The BTDF algorithm uses the basic weight and bursty weight of words in Weibo text to identify burst features.By analyzing the short-term and long-term pre-information of Weibo text,the pseudo-burst features are filtered.Combining the identified burst features with the results of the current time slice topic discovery,a burst topic is detected.Experimental results show that our proposed BTDF algorithm can accurately and effectively detect unexpected topics.(3)This thesis proposes an algorithm for mining the emergent topics and discovering the evolution rules to realize the mining of topic features in different time windows of the bursty topics.Topic characteristics usually include characteristics such as topical heat and keywords.By mining changes in the number of Weibo comments,likes,and blog posts in the topic cycle,the evolution of the heat is discovered.Text keywords are extracted by combining the semantic relevance and co-occurrence relations of words in Weibo text.By showing the changes of topic features in different dimensions of a topic with time series,the discovery of evolution laws was realized.(4)This thesis designs and implements a social network national security bursty topic detection,mining and evolution law discovery system.At the same time,the feasibility and effectiveness of the proposed algorithm are verified.The system consists of four modules:a data collection and deep feature extraction module,a bursty topic detection module,a bursty topic mining and evolution law discovery module,and a system display module.