Research On User Role Partition And Recognition In Social Network

With the widespread popularity of social media and online social networks,the process of forming and sharing opinions has fundamentally changed.Online social networking platforms enable users to communicate with each other.Through these online social platforms,users can easily express their opinions and meet other people.Due to differences in shared attributes such as hobbies,social relationships and geographic location,various communities have been formed.As social media is increasingly shaping opinions,many social problems have arisen.Positive opinions will have a major impact on social development,while negative opinions will have a serious impact on social stability without control and guidance.Understanding how emotions spread through social networks and identifying users with high emotion influence are helpful to the positive guidance of online public opinions,which is of great significance to national security and social harmony,and it also has an immeasurable market value.However,the underlying mechanism of emotion contagion in social networks has not been well studied.On the one hand,few studies have analyzed the emotions that users share in social networks,and it is difficult to quantify the role played by different users in the process of emotion contagion.On the other hand,the relationship between the environment in social networks and the emotions of users is not clear.To address these problems,this thesis utilizes sociology and artificial intelligence technology and other cross-domain knowledge to study the potential mechanism of emotion contagion in social networks.The research of this thesis can be divided into the following four points:(1)We propose an emotional community detection algorithm based on the emotional preference similarity.Because of the existence of the phenomenon of emotion flocking in social networks,it has been discovered that users with the same attributes and emotional preferences are clustered together.Therefore,the emotional preferences of users in the network are quantified to calculate the emotional preference similarity among users.Subsequently,the Louvain algorithm was improved by using the emotional preference similarity,and propose an emotional community detection algorithm based on the emotional preference similarity.Discover the emotional communities from social networks.(2)We propose an emotion role mining model based on multi-view ensemble learning(ERM-ME).First,three different types of emotional features are proposed to quantify the emotion influence of users in social networks and the ability of cross-community emotional transmission.Then,we construct different emotional features,and employ the local fusion strategy to obtain meta-classifiers with global view.In global fusion stage,the output of the meta-classifiers are integrated through a weighted voting scheme based on the accuracy rate.(3)We propose a surrounding-aware emotion inference model(SAEI).A memory-based attention model is proposed.By analyzing the time-evolving influence and surrounding influence in social networks,an surrounding attention network and a time-evolving attention network are constructed,and the user's emotions establish connections with his neighbors' emotional states.And time interval sequence is introduced to solve the problem of irregular time series data in social network.(4)We conduct experiments to demonstrate the feasibility of the proposed algorithms and models.Experiments prove the effectiveness of the emotional community detection algorithm on the task of emotional community detection,the superiority of ERM-ME on the task of emotion role recognition and the excellent performance of SAEI on the task of emotion prediction.In the case study,the emotion role identified by the ERM-ME model is compared with traditional social role identification methods,which proves that the emotion influence of emotion role in social networks and the ability to spread emotions across communities are indeed better.The empirical analysis proves that emotion leader and emotion mediator play different roles in social network,there is a need for further division.The ablation experiment of the SAEI model prove the effectiveness of surrounding influence and time-evolving influence for emotion inference.