Research On Evolution Mechanism Of Knowledge Collaboration Network Based On Deep Learning

With the mature and wide application of information technology,knowledge collaboration shows a network trend of large-scale group participation and domain crossing.Existing studies have shown that knowledge collaboration networks often exhibit group efficacy beyond the superposition of individual efficacy,and the formation of such group efficacy is often related to the nontrivial network structure driven by the interaction between individuals,stimulating the exploration of underlying mechanism for knowledge collaboration network structures in the academic circle.However,the existing studies are still insufficient in the aspect of depth and systematicness.On the one hand,most studies focus on the static structure of the network,but pay little attention to the evolution characteristics of the network.On the other hand,existing research approaches often adopt the combination of data analysis and agent-based modeling,which divorced from the real network to a certain extent and may inevitably affect the accuracy of the research results.Accordingly,this thesis analyzes the structural and evolutionary characteristics of knowledge collaboration networks in two different fields,and then constructs a deep-learning framework to detect the underlying mechanism of structural evolution of the real networks.The main findings are as follows.(1)Taking the fields of open-source software development and scientific collaboration as examples,thesis analyzes the structure and evolution of corresponding knowledge collaboration networks,and finds that such two networks exhibit an evolutionary process marked by structural changes of the giant component.Specifically,the giant component evolves from a cluster to a chain structure with multiple communities,then further evolves into a modular small-world network.In addition,through the analysis of the characteristics of the connection relationship between the two networks,it is found that the two parties who establish inner-community connection have similar professional background knowledge,while the two parties who establish inter-community connection have different professional background knowledge.Accordingly,the evolution process of network structure may be the result of the interaction of homophily and heterophily.(2)On the basis of(1),this thesis proposes a two-stage deep-learning model framework to analyze the individual interaction mechanisms that drive the evolution of network structures.In the first stage of the model,the graph embedding method is used to obtain the representation vector of node structure and non-structure attributes.In the second stage,through the supervised learning method,we can capture the connecting edge rule between the representation vector pairs and transform mechanism exploration issues into relationship prediction problems.Based on real network data,this thesis constructs three models based on the proposed model framework to examine the impact of homophily and heterophily on the formation of network structure.The results show that when only considering structural factors,it is not possible to accurately predict the connectivity within or between communities;after incorporating the homophily mechanism,the accuracy of inner-community edge prediction significantly increases,while there is no significant change in the accuracy of inter-community edge prediction;further incorporating the heterophily mechanism,the accuracy of inner-community(inter-community)prediction reaches a high level.In addition,considering both homophily mechanism and heterophily mechanism,the predicted network topology maintains a high degree of consistency with the real network.The above results indicate that the homophily mechanism and heterophily mechanism are the main driving factors for the evolution of the two network structures examined in this thesis.