CS224W Final Report Emergence of Global Status Hierarchy in Social Networks

Social network analysis provides insights into a wide range of social computing applications. The fundamental element of social network is the relations between people. By analyzing the relations between people in a network, we can reveal interesting structures and useful insights about the network. Relations between people in real world networks often exhibit two prominent features: (1) it is a mixture of positive (e.g. supporting) and negative (e.g. opposing) interactions; (2) it evolves through time. While many people have studied relations between people and network structure, most researches on social networks have focused on positive interactions between people. Only recently people have taken negative interactions into account. Moreover, few studies have been presented on the evolution of signed networks. In addition, many works are often limited to theoretical analysis under strong assumptions. In this course project, we address the above problems by analyzing the evolution of real world signed network and building models for signed network evolution. Interesting insights are discussed by analyzing real world network and evaluating our models based on it. In the first part of the project, we analyzed how status hierarchy evolves and converges to a true global status hierarchy in real world signed network. We designed algorithms to compute status hierarchy of a network. Given an evolving signed network, we assumed that the status hierarchy, which was computed in the final state, represented the true underlying global status hierarchy. We used normalized discounted cumulative gain score, which was widely used in information retrieval community, to analyze how the status hierarchy followed the global status hierarchy over time. Our analysis shows that the signed edges of Epinions network follow the global status hierarchy more and more. In the second part of the project, we proposed different models of network evolution. The models are based on local status information (i.e. how the two nodes involved in the link formation process give and receive signed edges) as well as global status information (i.e. the status hierarchy of the network at the time of link formation, and the community’s overall link creation behavior). We evaluated our models on real world network and the experiment results show that users create signed links based mostly on local information, rather than global information. The rest of this paper is organized as follows: Section 2 introduces related work; Section 3 presents the evolution analysis algorithms and status prediction models; Section 4 investigates the experiment and simulation results for network evolution and status prediction. Section 5 highlights the key results and concludes the paper.