GA-TVRC: A Novel Relational Time Varying Classifier to Extract Temporal Information Using Genetic Algorithms

Abstract

Almost all networks in real world evolve over time, and analysis of these temporal changes may help in understanding or explanation of some properties or processes of a network. This paper presents GA-TVRC, a novel Relational Time Varying Classifier which uses Genetic Algorithms to extract temporal information. GA-TVRC uses Evolutionary Strategies to optimize the influence of each previous time period on classification of new nodes. A Relational Bayesian Classifier (RBC) that is proposed by Neville et.al. [3] is utilized to compute the fitness function. The performance of GA-TVRC is compared with both the RBC, which ignores the time effect and the time varying relational classifier (TVRC) that is proposed by Sharan and Neville [20]. TVRC improves the RBC by taking the time effect into account using different predetermined weights. According to the experiments on two real world datasets, GA-TVRC extracts time effect better than the previous methods and improves the classification performance by up to 5% compared to TVRC and up to 10% compared to RBC.