Mining And Analyzing Multivariate Collaborative Relationships In Academic Networks

Scientific collaboration has become one of the most common and significant academic be-haviours.It is also the most fundamental organization form of research patterns.Research on such kind of relationships can improve the understanding of academic society,enhance the effi-ciency of scientific research,optimize the allocation of scientific research,and promote scientific progress.Team-based multivariate collaboration has been listed as a significant scientific re-search pattern in making breakthroughs of main scientific issues as well as engineering projects.As the hot research spot of network science.organization science,society science,etc.,mining multivariate collaborative relationships in academia still remain many challenges.There lacks a universal algebra model for collaborative teamwork,so that current studies cannot be integrated.Also,current algorithms generally lack the ability to handle the large-scale and the contagious distribution of academic networks.Meanwhile,the increasing scale and complexity of academic networks make it difficult to cluster collaborative relationships as well as recognize outlier rela-tionships.Based on the research methods and theories of computational social science,network sci-ence,data science,this thesis mainly focuses on the issues brought by the high dynamic of aca-demic society,the complexity of academic networks,and the heterogeneity of academic entities.Specifically,this thesis aims at solving the difficulties of formulating multivariate collaboration relationships,identifying team-based collaboration relationships,enhancing the generality of multivariate relationships clustering,and recognizing abnormal multivariate relationships.This thesis correspondingly proposes a Liebig' s Barrel based team collaboration relationships formu-lation.collaboration intensity based team recognition,k connected subgraph based multivariate relationships clustering,and motif similarity based outlier multivariate relationships detection.By mining and analyzing multivariate collaboration relationships in academic society,this thesis makes the following contributions:1.Team collaboration relationships formulation.Focus on the deficiency of general algebra model of collaborative teams,this thesis proposes a Liebig's barrel based algebra description for collaborative teams and analyzes the key factors of teamwork output.By mining the data from computer science and physics disciplines,this thesis find out that "anti-cask" effect exists in academic collaborative teams.That is.the quality of teamwork output mainly depends on the member with highest ability.Besides,a team with staircase structure generally performs better.Based on the achieved conclusions,policy makers can optimize collaboration team structures to improve scientific output.2.Team collaboration relationships recognition.Focus on the difficulty in recognizing col-laborative teams,this thesis proposes an algorithm based on network structure and collaboration intensity.By firstly refining the collaboration network by collaboration intensity,and then quan-tifying leadership and influence of team members,the collaborative teams with "core+extended"staircase structures can be recognized.Experimental results indicates that this method can rec-ognize fine-grained collaborative teams and describe team structures.3.Multivariate collaborative relationships clustering.Focus on the insufficiency of cluster-ing multivariate collaborative relationships in large-scale networks,this thesis proposes a cluster-ing algorithm based on k-connected subgraphs network partition.Based on the network connec-tivity theory and matrix perturbation theory,this thesis proves that after using the k-connected subgraph to partition the networks,the eigenvalues of adjacency matrix are not or just slightly effected.The proposed method is applicable for large-scale academic networks as well as higher-order motif structures.4.Outlier multivariate collaborative relationships detection.Focus on the diversification and higher-order abnormal relationships,this thesis proposes an outlier multivariate collaborative relationships detection algorithm.Based on the query conditions given by users,this algorithm can search related query results.By ranking the motif similarities within the query result set,the outlier multivariate collaborative relationships can be detected.Experimental results show that this algorithm is sensitive to the users' queries while not to the similarity measures.