On Dynamic Models Of Collaboration Networks Driven By Homophily And Heterophily

In reality, the collaboration networks (e.g. scientific collaboration networks and open source software development networks), which are composed of adaptive individuals, have incomparable advantages in solving complex social and economic problems. Many empirical studies indicate that the complex-problem-solving ability of these collaboration networks comes from the members’coordination and the corresponding network structures and evolutionary processes, and these collaboration networks often have significant community and small-world structures. Although the characteristics of community and small-world in collaboration networks have been intensively discussed in the past two decades, our understanding on the generative mechanisms and evolutionary processes of multi-communal small-world structures in collaboration networks should still be deepened. Especially, existing works usually attribute the formation of multi-communal small-world network to the combination of two mechanisms in the perspective of social capital (i.e. structural embeddedness and structural holes). However, other empirical studies have also illustrated that the formation of collaboration relationships is also related to the similarity (or heterogeneity) of individual trait. This dissertation emphasizes the role of individual interaction based on trait similarity (or heterogeneity) (i.e. homophily, heterophily, and social learning) should not be overlooked, arguing that it may be a critical force for the formation of multi-communal small-world structure in collaboration networks. Therefore, by reviewing existing works, this dissertation adopts agent-based modeling and data analysis to investigate the structure and evolution of collaboration networks driven by the above-mentioned three mechanisms.Firstly, the dissertation tests the effects of structural embeddedness and structural holes on network structure formation through an edge-rewiring model; and the results reveal that the generated networks have neither a stable small-world state nor significant community structures. Furthermore, the dissertation presents an edge-rewiring model driven by homophily, heterophily, and social learning. The simulation results indicate homophily is a clustering mechanism to form communities of homogenous agents (i.e. individuals), heterophily induces the formation of inter-community links, and social learning impels trait convergence of similar agents which causes edge density increase in each community so that the community structure is reinforced. Hence, under the combinatorial effects of the three mechanisms (especially under a certain combination of homophily and heterophily), the generated network can evolve into a stable small-world state with community structure. These results not only compensate for the lack of comprehensive investigations for these mechanisms in existing works, but also clarify the formation process of community structure and small-world state in collaboration networks from the perspective of individual interaction based on trait similarity (or heterogeneity).Secondly, on the basis of identifying the effects of homophily, heterophily, and social learning for the formation of network structure, this dissertation further presents a dynamic model driven by homophily and heterophily to investigate the evolutionary process of generated networks under the scenario of scaling-up networks. The simulation results indicate the generated network gradually evolves from segregation state to a "core-periphery" structure, and such evolutionary process can be divided into three stages in turn (i.e. segregation network, chained-community structure, and multi-communal small-world network) according to the structural changes of the "core"(i.e. the giant component). This result may be a good complement for theoretical studies of collaboration network evolution, which can also contribute to the development of related theories.Finally, this dissertation selects three real networks as examples (i.e. the co-authorship networks in "evolution of cooperation" and "complex network" fields, and email communication network of participants in Linux kernel development) to further explore the structure and evolutionary process of collaboration networks. Analysis results show that all the three networks have a three-stage evolutionary process according to structural changes of the giant component, and finally evolve into modular and hierarchical small-world networks. This study may be complementary to existing empirical works, which is helpful to deepen our understandings on the evolutionary mode of these collaboration networks. Furthermore, such evolutionary process is consistent with the findings in numerical experiments, which implies the model presented in this dissertation can reflect the actual evolution of collaboration networks. Hence, the individual interaction based on trait similarity (or heterogeneity) may be the driving force of multi-communal small-world network formation in collaboration networks, providing theoretical supports for managerial practice.