Social Network Analysis using Cultural Algorithms and its Variants

Finding relationships between social entities and discovering the underlying structures of networks are fundamental tasks for analyzing social networks. In recent years, various methods have been suggested to study these networks efficiently, however, due to the dynamic and complex nature that these networks have, a lot of open problems still exist in the field. The aim of this research is to propose an integrated computational model to study the structure and behavior of the complex social network. The focus of this research work is on two major classic problems in the field which are called community detection and link prediction. Moreover, a problem of population adaptation through knowledge migration in real-life social systems has been identified to model and study through the proposed method. To the best of our knowledge, this is the first work in the field which is exploring this concept through this approach. In this research, a new adaptive knowledge-based evolutionary framework is defined to investigate the structure of social networks by adopting a multi-population cultural algorithm. The core of the model is designed based on a unique community-oriented approach to estimate the existence of a relationship between social entities in the network. In each evolutionary cycle, the normative knowledge is shaped through the extraction of the topological knowledge from the structure of the network. This source of knowledge is utilized for the various network analysis tasks such as estimating the quality of relation between social entities, related studies regarding the link prediction, population adaption, and knowledge formation. The main contributions of this work can be summarized in introducing a novel method to define, extract and represent different sources of knowledge from a snapshot of a given network to determine the range of the optimal solution, and building a probability matrix to show the quality of relations between pairs of actors in the system. Introducing a new similarity metric, utilizing the prior knowledge in dynamic social network analysis and study the co-evolution of societies in a case of individual migration are another major contributions of this work. According to the obtained results, utilizing the proposed approach in community detection problem can reduce the search space size by 80%. It also can improve the accuracy of the search process in high dense networks by up to 30% compared with the other well-known methods. Addressing the link prediction problem through the proposed approach also can reach the comparable results with other methods and predict the next state of the system with a notably high accuracy. In addition, the obtained results from the study of population adaption through knowledge migration indicate that population with prior knowledge about an environment can adapt themselves to the new environment faster than the ones who do not have this knowledge if the level of changes between the two environments is less than 25%. Therefore, utilizing this approach in dynamic social network analysis can reduce the search time and space significantly (up to above 90%), if the snapshots of the system are taken when the level of changes in the network structure is within 25%. In summary, the experimental results indicate that this knowledge-based approach is capable of exploring the evolution and structure of the network with the high level of accuracy while it improves the performance by reducing the search space and processing time