Tuning Latent Dirichlet Allocation Parameters using Ant Colony Optimization

Abstract

Latent Dirichlet Allocation is a famous and commonly used model used to find hidden topic and apply in many text analysis research. To improve the performance of LDA, two Dirichlet prior parameters, namely the α and the β, that has an effect on the performance of the system are utilized. Accordingly, they must be set to an appropriate value. Ant colony optimization has the ability to solve the computational problem by adding parameters tuning. Thus, we proposed to implement an approach to find the optimal parameters α and β for LDA by using Ant colony optimization. An evaluation using dataset from the UCI (KOS, NIPS, ENRON) that are the standards for estimating topic model was conducted. The results of the experiment show that LDA, which has tuning parameters by ACO has better performance when it is evaluated by perplexity score.