Social networks are an ubiquitous element of our daily experience. A question that naturally arises in the operation of such networks is whether they can be controlled. With domains ranging from financial markets to extremism avoidance, the study of control of opinion in social networks is extremely relevant in modern social media culture. Automated control of the flow of information in large-scale non-deterministic social networks is a complex problem requiring both a search for the optimal configuration of connections to the network, and a behaviour that determines the required control signals. This thesis formalizes the Network Control Problem (NCP) as a means of describing the field of diverse social network control problems. To date, problems that can be described as NCP examples deal primarily with the configuration rather than behavioural component. The θ-Consensus Avoidance Problem (θ-CAP) is defined in this thesis as a novel NCP which has the objective of avoiding consensus in a social network of agents. It is an important problem representing the avoidance of extreme views that may lead to extreme behaviour, such as bubble or panic events in a socially-connected market. The θ-CAP is intended as a foundational benchmark problem in the NCP domain. Experimentation is developed to demonstrate the utility of the θ-CAP as a practical benchmark problem with scalable difficulty parameters. A number of heuristic and metaheuristic implementations of both configuration and behaviour are compared in this thesis. Special consideration is given to the application of evolutionary neurocontrollers toward learning to optimize the behaviour component. Analysis indicates a variety of conditions that affect the quality of given approaches, including network structure, instance difficulty parameters, and variations among control approaches. Consistent, observable trends were determined in the solution space that were verfied across behaviours, configurations, network structures, and objectives.