Mechanisms Of Multi-Stage Oriented Emotion Propagation In Online Social Networks

The explosive growth of Internet interaction information has a profound impact on the politics,economy,culture and education of our society,as well as personal work,life and entertainment.Emotion propagation not only in real life,but also implicitly affects people’s interactive behaviors through Internet.The study of emotion propagation mechanisms is an indispensable part of interaction theory,and can assist public opinion monitoring and public opinion guiding.As a network form bred by the Internet,social networks are often oriented by individual interests.The social network events and caused emotions usually only spread locally,which generally do not cause a comprehensive social impact.However,once social network events break out,they will form uncontrollable Internet events,then affect social stability and national security,which should be avoided.In addition,the emotions triggered by social network events are mainly transmitted through text,voice,image and so on.Here,this thesis focuses on the emotion propagation in text-based social networks,which can assist the social network public opinion monitoring and public opinion guidance to avoid triggering Internet events.However,the traditional emotion propagation research usually ignores that the emotion propagation process and the event development process are two intertwined dynamic processes,and fails to comprehensively and carefully explore the multi-stage mechanisms and internal driving factors.Inspired by the relevant theories in social sciences,this thesis following the development law of social network events(which could be divided into four stages of initiation,expansion,variation,and deposition),innovatively proposed multi-stage emotion propagation mechanisms(i.e.,macro mechanisms),i.e.,emotional arousal,emotional diffusion,emotional fermentation,and fading/outbreak.We progressively explore the patterns and mechanisms of the four stages(i.e.,micro mechanisms),and propose four core tasks,including revealing social emotions,modeling individual emotion diffusion,modeling group emotion spreading,and explaining public sentiment evolution spikes.The main contributions of this thesis are summarized as follows:1)A study on emotion contagion oriented social emotion mining.This study attempts to mine the social emotions(i.e.,the first emotional reactions of public)triggered by social network events,which is the first stage of emotion propagation process.We extend the social emotion mining methods,which are only applicable to news websites,to text-based social networks(e.g.,Twitter),analyze the impact of emotional contagion phenomenon on social emotion mining,and mitigate the problem of under-informed and noisy text content by introducing community homogeneity.We propose the CommunityEnhanced Social Emotion Mining Model(CESEM),which can not only mine social emotions in text-based social networks,but also solve the problem of under-informed and noisy text content.Experiments show that CESEM improves the accuracy of social emotion mining by introducing emotion contagion mechanism and community homogeneity.2)A study on multi-factor oriented individual emotion diffusion modeling.The social emotion triggered by events in study(1)will further spread dramatically among individuals in social networks.The purpose of this study is to reveal the individual emotion diffusion mechanism,which is the main concern of traditional emotion propagation research.However,individual emotions are influenced by a variety of intrinsic and extrinsic factors,and there is a mismatch between these various factors.We propose the Multi-Factor Emotion Diffusion Model(MFEDM),which reveals three types of factors,including the diffusion variability of events,the slowness change of emotions,and emotion contagion.And we integrate individual personality into the emotion contagion factor to refine the diffusion strength,thus solving the mismatch problem.The experimental results also demonstrate that this model can significantly improve the performance of emotion diffusion prediction.3)A study on community oriented group emotion diffusion modeling.With the continuous dissemination of emotions in the study(2),public emotions will be diversified.Individuals with the same opinions will form groups(emotional identification),and groups with different opinions will collide violently(emotional conflict).The diffusion of group emotions will affect the division and reformation of communities.We propose the Emotion Diffusion-Oriented Community Detection Model(EDCD),which reveals the different emotion diffusion mechanisms within and between communities,and the impact of emotion diffusion information on community formation.Experiments show that this model effectively improves the performance of community detection,and can track emotion diffusion strengths within and between communities.4)A study on emotional identity oriented sentiment spike explanation.Public sentiment will disappear in the flood of information as time passes,but it may also erupt uncontrollably.Massive emotional spikes are the precursors of group events.This study aims to uncover the reasons for the sudden changes in public sentiment evolution.We propose the Emotional Identity-Oriented Emerging Topic Mining Model(EIETM),which employs three independent decision distributions(corresponding to three types of emotional identities)to automatically learn the potential orientation pattern,and explores the co-occurrence relationship between background and emerging topics.Experiments show that this model can not only effectively improve the quality of emerging topics,but also more accurately find the true reasons for sentiment spikes.The above four stages are complementary,although independent but continuous,making the study of emotion propagation an organic whole with significant theoretical and practical application value.