Research On Influence And Dynamic Evolution Analysis Of Internet Public Opinion Events

Nowadays in the networked and big-data era,social media has already become the main platform for the generation and propagation of public opinion events.There exist such significant challenges with national,societal,and enterprise security implications as event-driven online-offline real-time interactions,and rapid unconventional information propagation and contagion,among others.As a result,from a practical standpoint,influence evaluation and tracking of public opinions play an essential role in online monitoring and early warning concerning public opinion networks.Under the backgrounds,this dissertation focuses on the processing of multi-source heterogeneous data in the context of public opinion events,the main work of this dissertation are summarized below.Firstly,we have proposed an event impact assessment scheme for public opinions embedded in multi-source heterogeneous networks.The current assessment scheme has several major deficiencies.They typically ignore the multi-dimensional,and multisource nature of the information space and do not provide a comprehensive and dynamic assessment framework.In our work,we have proposed an assessment scheme that integrates cross-platform and heterogeneous information sources.It is designed to explicitly consider news sites,blog sites,microblogging sites,and other data sources.In addition to information integration,it features results from affective computing and sensitive information mining,and utilizes a multi-expert fuzzy evaluation method to assign weights to and integrate various factors.Experimental results indicate that the proposed assessment scheme delivers meaningful evaluation results and outperforms those based on single data sources.Secondly,we study the extraction and dectection methods of entity-words and its morphs expansion in the public opinion events.Motivated by the fact that the event describe keywords is dynamic changing in public opinion event evolution,we propose a double-layers model for the events' entity-words extraction,and a query expansionbased method by integrating the users' social relationship networks and word variations to recognize the Chinese morphs in the event evolution process,then the expansion words of the methods is verified by event-oriented data collection.Experimental results show that the proposed method provides an effective word extension and variation strategy,enhancing event-oriented data collection,especially with censored events keywords.Thirdly,we propose a heterogeneous community evolution analysis method based on a hybrid-model Dirichlet process,taking into account the unknown number of the communities to be identified,and the continuous and dynamic changing nature of event themes.This proposed method models microblogging articles as nodes,and constructs heterogeneous information networks for public opinion events using words as attributes.Sub-network sequences are used to capture different time periods.Based on the constructed networks,dynamic analyses are performed to study evolution of user communities,threads and posts.Experimental results show that this analysis method can better characterize the dynamic evolution of public opinions and community activities.The method has effectively solved the problems of communitys' count selflearning,community structure smooth evolution and community-topic collaborative detection.