Adaptive Firefly Optimization on Reducing High Dimensional Weighted Word Affinity Graph

Abstract

Document analysis and retrieval system can best define an efficient information retrieval system. Among various processing stages in a document analysis and retrieval system, feature descriptors at processing volume limit require more importance while developing the system. This is mainly because of the increase in probability of getting high dimensional semantic description. This increases the vitality of opting a robust dimensionality reduction method for our retrieval system. Principle Component Analysis (PCA), Independent Component Analysis (ICA), etc are the most popular dimensionality reduction methods. However, they are highly complex while handling nonlinear data with multiple characteristics. Optimization algorithms can be a good alternative for the traditional methods. In fact, classical optimization algorithms such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), etc have been widely applied. However, the data handling remains inefficient under current data exploding scenario. In our previous work, we have exploited Firefly Algorithm (FA) to solve the optimization problem. Due to parameter selection dilemma in traditional FA, this paper concentrates on using Adaptive Firefly Algorithm (AFA). AFA adaptively varies step search of solutions and hence improves the convergence rate of the algorithm. As a result, near – optimal solution can be obtained qualitatively. We further recommend the dimensionality reduction method to handle weighted word affinity graph to improve the retrieval efficiency.