Cooperation and competition are two typical interactional relationships in natural and engineering networked systems. Some interesting collective behaviors can emerge through local interactions within the networked systems. This paper considers a bipartite consensus for a group of agents interacting both cooperatively and competitively. Meanwhile, the intrinsic nonlinear dynamics of all the agents are unknown for the neighboring agents. Thus, the interaction network, which we call a coopetition network, is firstly represented by a signed graph. Then, some adaptive estimate laws are proposed for the unknown dynamics. Finally, an adaptive tracking control is designed for each agent to reach a bipartite consensus on the state of the leader. Some sufficient conditions are derived for bipartite consensus of multi-agent systems with the help of a structural balance theory. Simulation results are provided to demonstrate the bipartite consensus formation.