| This thesis investigates the assembly of ad hoc project teams through three interrelated studies. The first study investigates the assembly mechanisms of project teams by taking a bipartite network perspective. Individuals and projects are modeled as two types of nodes and team membership as relations between the nodes. Using Exponential Random Graph Models (ERGM/p*), the study examines players' team combat collaboration in a Massively Multiplayer Online Role-Playing Game (MMORPG). Empirical results show that individuals are motivated to join ad hoc teams to complete difficult projects, but not projects with long durations. It is also found that individuals tend to collaborate with specific teammates who have complementary skills, those who have similar age or skill level, and those who have the same organizational affiliation.;The second study explores the impact of composition factors such as expertise diversity and demographic diversity, and structural factors including intra- and inter-team relations on short-term, long-term, and negative team performance. Analysis results show that high short-term performance is related to high expertise diversity, low age diversity, low intra-team network density, and low cosmopolitan level; high long-term performance is related to low organization diversity and high intra-team network density, centralization and triangles; and high negative performance (less death) is related to high expertise diversity, low network density, and more triangles. Intra-team network density is also found to have interesting interaction effects with some composition factors. Two other research methods, including correspondence analysis and hypergraph models, are also introduced as better alternatives to study teams with overlapping members.;Built on the findings from the first two studies, the third study defines a team-assembly optimization problem, which includes three social factors, expertise diversity, attribute homophily, and social network effects, in the constraints. This study theoretically analyzes this optimization problem and proves the upper bound of the best solutions. An agent-based optimization model is proposed to solve this problem. Simulations results show that the solutions found using this model are over or close to 85% of the upper bound under various network structures and team sizes. It also out-performs another model built using greedy method. |
