Quantitative Method Research On The Influence And Susceptibility Of Social Network Users And Its Application

Social network analysis is a significant field within computer science.The study of information dissemination dynamics in social networks holds substantial theoretical and practical importance.Understanding these dynamics can help address key issues such as polarization,information cocoons,rumor control,identifying top spreading capacity spreaders,and friend recommendations.This dissertation is an interdisciplinary application within computer science.Utilizing information dissemination models and machine learning algorithms,we analyze the patterns and mechanisms of information spreading in social networks through both simulation and empirical studies.The research focuses on identifying vital nodes,analyzing the impact of user behavior characteristics on information dissemination,predicting user interaction behaviors,and forecasting top spreading capacity spreaders.In the study of information spreading dynamics in social networks,previous research has primarily focused on whether users have high spreading capability in the later stages of dissemination.However,timeliness is also crucial in information spreading.For instance,in achieving effective short-term promotion of new products or marketing of new movies,identifying users with high spreading capability in the early stages is of significant practical importance.Moreover,a few studies published in top journals like Science have found that the behavioral traits of user influence and susceptibility help to understand the information spreading process from a micro perspective.These two user behavioral traits refer to abilities related to the users themselves,independent of the network structure.However,up to now,research on information dissemination dynamics in social networks has focused more on the impact of network structural characteristics on information spreading.There has been less discussion on how user influence and susceptibility affect information dissemination,and no studies have proposed methods to quantify these two behavioral traits of users.To solve the above problems related to information dissemination dynamics.Firstly,chapter 3 of the dissertation introduces the concept of fast influencers and systematically analyzes the differences between fast influencers and users with high spreading capabilities in the later stages under the Susceptible-Infected-Removed spreading model,and proposes a local centrality index,called social capital,which can identify fast influencers effectively.This dissertation finds that there is an overlap between fast influencers and users with high spreading capabilities,meaning some users are both fast influencers and users with high spreading capabilities in later stages.The thesis focuses on the non-overlapping parts: fast-only influencers and later-only influencers and analyzes their differences both in spreading capabilities and network structures.The results show that as the spreading probability increases,users with high spreading capabilities in later stages need more time to surpass fast-only influencers.Furthermore,fast-only influencers also have stronger local network structures.By controlling network degree assortativity,this dissertation also finds that in more assortative networks,fast influencers are more likely to transition into users with high spreading capabilities in later stages.Then,chapter 4 of the dissertation proposes a coupled nonlinear algorithm that quantifies user influence and susceptibility using information propagation data,and the convergence of the proposed algorithm is analyzed from theoretical and numerical experiments,providing an unnecessary and sufficient convergence condition for the algorithm.Additionally,the robustness of the proposed algorithm is verified as well.The thesis applied the algorithm to four empirical spreading datasets collected from Weibo and Twitter social platforms to quantify the influence and susceptibility of users.The experiments find that in the empirical spreading networks constructed from the datasets,users’ outdegree and indegree distributions follow a power-law long-tail distribution,while users’ influence and susceptibility follow a bell-shaped distribution.Finally,to further understand user behavioral traits,chapter 5 of the dissertation discusses the roles of user influence and susceptibility in information propagation under different application scenarios.Under the assumption that users have two behavioral traits:influence and susceptibility,this dissertation selects high-degree users as seeds,uses an independent cascade spreading model to simulate the information spreading,and immunizes highly influential and highly susceptible users,respectively.The results show that immunizing highly susceptible users can decrease the final spread size more effectively.Then,this dissertation investigates the spreading probability prediction problem in the four empirical datasets,namely,the weighted network link prediction problem.Specifically,using the proposed algorithm,this dissertation quantifies users’ influence and susceptibility,and based on these two user behavioral traits,this dissertation constructs a link prediction metric to predict the spreading probability between users.The results show that the link prediction metric based on user behavioral traits outperforms traditional link prediction metrics based on common neighbors.Furthermore,using algorithms from computer science such as Random Forest,XGBoost,and Multilayer Perceptron,this dissertation constructs several machine learning models to identify top spreading capacity spreaders in the network based on user behavior traits and network structures.The results show that,compared to network structural features,the user behavior traits can identify top spreading capacity spreaders more effectively.This result also provides new insights into understanding the positions of top spreading capacity spreaders in the network from the perspective of user behavior traits.