Enterprise Social Network Based Team Formation

Large enterprises,especially large IT companies,form teams with different sizes and structures,to complete projects or tasks.With the growth of business scale,enterprises are seeking a more scientific and reasonable way to form teams,in order to achieve efficient operation,low project cost,high utilization of human resources and high project quality.In the age of the Internet,large software outsourcing companies are faced with new,fast-changing business needs.Immutable and empirical approaches can not play its in-herent advantages in forming teams for projects of these companies,which leads to huge commercial value for team formation.Activity records generated by management soft-wares in enterprise reflect social relationships among members and contribute to the con-struction of enterprise social network.At the same time,project data can be leveraged to reproduce the needs of real projects along with its actual team members.These two conditions provide the possibility of forming collaborative teams for enterprise projects.Based on the research of enterprise social network and real project needs of a large software technology outsourcing service company,this paper studies team formation meth-ods and measurements in real application scenarios.Major contributions of this paper are listed as follows:1.Formalizing enterprise project team formation problem and proposing a new measurement.Based on the study of real project needs and team formation expe-rience,we formalize enterprise social network,tree-structured team specification,and enterprise social network based team formation problem.A new measurement Local Distance,is proposed,which reflects communication cost among team mem-bers and evaluates generated teams in a more reasonable way.2.Proposing three team formation algorithms and an interactive strategy.Two heuristic algorithms are proposed to solve tree-structured team formation problem,Top-down and Bottom-up team formation algorithm,which focus on optimizing local distance.Based on analysis of real world sub-teams patterns and complexity of proposed algorithms,an index-based algorithm and an interactive strategy are proposed to make team formation process faster and interactable.3.Conducting experiments on a real enterprise social network,verifying high ef-fectiveness and efficiency of proposed algorithms.Large-scale experiments are conducted on project data provided by a large IT company,which contains more than 70,000 real projects and 150,000 employees.It is verified by statistical ex-periments that Local Distance is a better than Glbal Distance as a measurement for generated project teams.Besides,proposed methods prove to surpasses three traditional algorithms in both effectiveness and efficiency.4.Designing and implementing an enterprise team formation system,TeamGen.An enterprise team formation system,TeamGen,was designed and implemented,which provides services for more than 150,000 employees,including enterprise so-cial network browsing and team formation.In summary,this thesis studies team formation problem based on real enterprise project data.It first formalizes the enterprise social network,tree-structured team forma-tion problem and proposes a new measurement for generated teams.Second,three team formation algorithms and an interactive strategy are proposed to solve the problem.Third,it is verified that proposed algorithms achieve better effectiveness and efficiency than tra-ditional algorithms by experiments of real project team simulations.Finally TeamGen,an enterprise team formation system,is currently serving in a large enterprise.