Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science

First Advisor

Ziad Kobti


Artificial Intelligence, Cultural Algorithms, Multi-Agent Based Simulation, Social Networks, Social Systems, Specialization




A social system is a patterned network of interrelationships that exist between individuals, institutions, and groups forming a coherent whole. Understanding the varying system outcomes for different decision-making processes selected under varying environment constraints in advance will aid in the realization the of best decision towards an effective outcome. One of the ways to increase system productivity is ‘Agent Specialization’. Also, the agents (individuals) who operate as generalists are most vulnerable to being replaced. Therefore, there is a need to focus on agent specialization to enhance the ability of an agent along with the evolution of an agent. Multi-Agent Based Simulation, a subfield of distributed AI, provides a technique to naturally describe a social system. To help improve decision-making intricacies of the agents to evolve and specialize, there is an increasing need to formulate an enhanced model of MABS. This thesis proposes a novel framework that exploits the benefits of social networks providing a decision support system for agent (individual) specialization by integrating the concept of ‘Positive Social Influence’ exerted by experts in the system. Consequently, the proposed framework assists the growth of agents by enabling the evolution of agent capabilities with the identification of suitable producer-agents using an evolutionary component (cultural algorithms). Enabling agent specialization and assisting the ability of the agents through capability evolution is anticipated to increase the productivity of the system. Evaluation of results shows the successful evolution of agent capabilities with the identification of suitable producer-agents in an optimized aspect (reduced operational cost and reduced distance cost) in comparison with exhaustive search, random search, and genetic algorithms and the improved degree of specialization of agents (increased dol values with a minimum of 3% increase to a maximum of 16.7% increase in comparison with standard genetic threshold model for varying agents and task number) in a given dynamic environment.