Research On Polarized Communities Search In Attributed Signed Network

With the widespread adoption of social media,the emergence of polarized communities alongside information dissemination and communication on these platforms has gained significant attention.Understanding the formation and evolution of polarized communities in social media has become a meaningful research direction in recent years.Polarized communities search aims to identify polarized communities that are highly relevant to a given pair of adversarial query nodes,such that the majority of nodes within each community form dense positive connections,while the majority of nodes between communities are connected by dense negative links.Existing studies have predominantly focused on network topology,however,nodes are often associated with a plethora of attributes,which provide rich and highly relevant auxiliary information.Accurately and effectively analyzing such structurally complex and information rich attributed signed network is crucial for capturing polarized communities precisely.Through summarizing and analyzing existing research efforts,it has been observed that current methodologies struggle to handle heterogeneous information from node attributes.Furthermore,effectively modeling the intricate relationship between network topology and node attributes in the identification of polarized communities remains a challenge.To address these issues,this thesis investigates the polarized communities search task in attributed signed networks by combining spectral graph theory and random walk mechanism.The major contributions are summarized as follows:First,we propose a method for Polarized Communities search in Attributed Signed network(PCAS)based on local spectral subspace.The method firstly performs a statistical analysis of the relationships between nodes in the attributed signed network at the attribute level to reveal auxiliary information that is different from the topology.Secondly,the topology and node attributes are effectively fused in an adaptive way to fully exploit the attributed signed network.Finally,the polarized communities search problem is transformed into a linear programming problem to solve an indicator vector supported by the query node,and discrete rounding of the indicator vector to capture the target polarized communities.Second,polarized communities search via Co-guided Random Walk in Attributed signed networks(Co RWA)method is proposed.Specifically,firstly,the method designs an attribute-based signed network to model the auxiliary relationships between nodes.Secondly,a new weight assignment mechanism is proposed based on the theory of structural balance in signed networks.This mechanism measures the reliability of edges to make full use of the a priori information hidden in the topology.A co-guided random walk mechanism in signed networks is further designed to model the relationship between network topology and node attributes in order to improve the search results.Finally,the Rayleigh entropy is defined to quantify and locate polarized communities.Experiments conducted on real-world datasets validate the effectiveness and practical application of the method proposed in this thesis.