A hybrid agent based virtual organization for studying knowledge evolution in social systems

Abstract

Social modeling applies computational methods and techniques to the analysis of social processes and human behavior. Cultural algorithms (CA's) are evolutionary systems which utilize agent technology and which supports any evolutionary strategy like genetic algorithm, evolutionary algorithm or swarm intelligence or ant algorithms. CA's have been used for modeling the evolution of complex social systems, for re-engineering rule based systems, for data mining, and for solving optimization problems. In the current study a cultural algorithm framework is used to model an Agent Based Virtual Organization (ABVO) for studying the dynamics of a social system at micro as well as macro level. Research gap exists in defining a concrete and systematic method for evaluating and validating Agent Based Social Systems (ABSS). Also the knowledge evolution process at micro and macro levels of an organization needs further exploration. The proposed CA is applied to the problem of multi-objective optimization (MOO) of classification rules. The evolutionary knowledge produced by the agents in creating the rules is accepted into the belief space of the CA and macro evolution takes place. The belief space in turn influences the agents in successive generations. The rules created by the individuals and the knowledge sources created during evolution provide a concrete method to evaluate both the individuals as well as the whole social system. The feasibility of the system has been tested on bench mark data sets and the results are encouraging.