Research On Hot Events Detection And Tracking Technology In Social Networks

Recently,with the rapid development of Internet technology,social networks,with a large amount of complex data and rich application scenarios,have penetrated into all walks of life,which provide a new data-driven perspective for the study of social events and patterns in addition to the traditional sociology.In order to effectively analyze social network data,and accurately detect hot events from massive social data and track key hot events correctly,the data mining techniques have emerged as new methods for social network data analytics.The detection and tracking of hot events in social networks are a key part of cybersecurity,and have become a hot research topic within the social media data mining in recent years.Typically,the detection and tracking techniques of hot events in social networks adopt the existing methods from the combination of traditional information retrieval and Internet technology.Through the analysis of user groups and a large amount of real-time data in social networks,the hot events along with their propagation and evolution can be effectively captured,thus,providing a strong support for network public opinion analytics.Since event detection,event propagation and event evolution in social networks are the key components of public opinion analytics,the effectiveness is closely related to the overall performance of hot event detection and tracking techniques.Therefore,the researches on event detection,event propagation and event evolution have important theoretical significance and practical value.Firstly,this dissertation systematically studies the relevant theoretical knowledge and key technologies of social networks and identifies some key limitations in the recent existing research works.Then,this dissertation proposes novel approaches to address these limitations focusing on event detection,event propagation and event evolution in social networks.The core contributions of this dissertation are listed as follows:(1)Most existing event detection methods in social networks cannot effectively filter out poor-quality posts,low-influence users and unpopular topics,and they lack identification ability of key posts and influential spreaders and lack the stability of the prior estimation algorithm of topic models,which leads to low accuracy and low efficiency in social event detection.To address the above issues,a Hypertext-Induced Topic Search(HITS)based Topic-Decision method(TD-HITS)and a Latent Dirichlet Allocation(LDA)based Three-Step model(TS-LDA)are proposed for event detection and influential spreader discovery in online social networks.Specifically,TD-HITS is firstly proposed to pre-process social network data,detect the number of topics as well as identify associated key posts in a large number of posts,and select high-quality posts,high-influence users and hot topics to improve the accuracy and efficiency of event detection.Then,TS-LDA is developed to automatically identify influential spreaders of hot event topics based on both posts and user's information according to an authority value and a minimum distance of posts,with which the prior parameters of LDA topic model can be determined by the number of key posts,thus,further improving the accuracy and efficiency of event detection.Next,key users in hot topics are identified according to the user's centrality and a minimum distance provided by TD-HITS.Meanwhile,influential spreaders are then identified in hot events combined with the activity and local features of key users.Finally,the experimental results demonstrate the effectiveness of our proposed methods in event detection and influential spreaders identification using a Twitter dataset.(2)Existing methods of influence maximization in social networks suffers from several limitations,such as low accuracy,low efficiency and narrow dissemination scope due to the neglect of the interest of influential spreaders,the popularity of topics and the identification of a suitable number of influential spreaders under popular topics.Thus,a high-influence greed maximization model based on a user's interest-topic model is proposed.Specifically,the user's interest-topic model based on the LDA topic model can process social network data to get the distribution of users' interest topics and topic sensitivity.Then,all posts and users in online social networks are pre-processed according to the user's topic sensitivity,and a subset of high-influence users under hot topics is obtained to improve the accuracy and efficiency of interest discovery of influential spreaders.Moreover,short-text posts in online social networks are clustered and the appropriate number of clusters is determined by a topic decision graph,which integrates the short-text posts in each cluster into a single post document to form a long-text post document and calculates the user's interest distribution under each hot topic using the proposed user's interest topic model,which further enhancing the user's influence under specific hot topics and improving the accuracy and efficiency of interest discovery for influential spreaders.Besides,posts and users in each hot topic are assigned with different weights to characterize their importance according to the HITS algorithm,so that each hot topic can be represented by multiple posts and users.Furthermore,a multi-prototype interest community detection model is proposed to identify user's interest communities,which can improve the accuracy and efficiency of identifying an optimal number of influential spreaders in hot topics.Moreover,the greed maximization model based on user's interest topic is used to rapidly determine the appropriate number of influential spreaders under hot topics,which can improve the accuracy and efficiency of the identification of influential spreaders under hot event topics with a wider scope of influential dissemination.Finally,the experimental results show that the proposed models are superior to the benchmarking models,and effectively and accurately maximize the influence of hot events.(3)The existing information dissemination models in social networks lack the selective dissemination according to the popularity of user's interests,fail to identify influential spreaders,and unable to learn any experience from the previous event dissemination process.As a result,most existing methods experience low accuracy,low efficiency and narrow dissemination range in continuous information dissemination.To resolve the aforementioned drawbacks,a user's interest popularity-based personalized event propagation model is proposed.Firstly,the user's interests are identified by the combinations of HITS algorithm,user interest topic model and topic decision graph to obtain their topic popularity for improving the accuracy and efficiency of the event propagation model.Secondly,the learning process is added to the event propagation process,and the existing events are used to disseminate information and user's interest information.Then,a personalized event-propagation model is created and updated through the joint calculation of user's interest-topic similarity judgment model,authoritative degree calculation model and event topic similarity analysis model,which improves the accuracy of continuous event propagation.Besides,the key role of influential spreaders and the characteristics of events are described in the dissemination process of events through the experience set,which can further improve the accuracy,efficiency and scope of events dissemination model.Finally,the experimental results verify the accuracy and efficiency of the proposed personalized event propagation model.(4)The existing event evolution methods in social network lack the tracking and identification ability of post's influence,user's influence and the dynamic change of user's interests.Moreover,the low detection rate of new and old hot events,the poor ability of event evolution control,and the lack of interest evolution of influential spreaders lead to the difficulties of tracking the evolution process of hot events efficiently and accurately as well as.To mitigate these challenges,a user interest evolution-based event evolution model is proposed.Specifically,a hot-topic clustering algorithm and the user's interest community detection algorithm are used to identify the influential spreader's interest community under the hot topics.Secondly,a novel method based on the user's network topology and user's interaction network structure is proposed to capture dynamic user influence.It can dynamically track the changing process of user's interest labels in social network interaction based on the Label Propagation Algorithm(LPA),and it also can identify the interest evolution process of influential spreaders efficiently.Then,a Term Frequency–Inverse Document Frequency(TF-IDF)-based keyword extraction method is used to extract the hot keywords of each topic efficiently and accurately,and the cosine distance between keywords of each hot event topic is calculated to judge whether it is a new event or the evolution of existing hot events,which can further improve the accuracy and efficiency of identifying new and old hot events in the process of event evolution.Finally,the experimental results demonstrated that the proposed approach is superior to the benchmarking methods in terms of efficiency and accuracy in identifying the new and old hot events in the event evolution process and the interest evolution process of influential spreaders.