There are existing implementations of Latent Dirichlet Allocation (LDA) algorithm as a semantic classifier to arrange the data for efficient retrieval. However, the problem of learning or inferencing the posterior distribution of the algorithm is trivial. Inferencing directly the prior distribution could lead to time taken to increase exponentially. It is due to the coupling of the hyperparameters. Several inference algorithms have been implemented together with LDA to solve this issue. The inference algorithm used in this research work is Gibbs sampling. Research using Gibbs sampling shows promising results in comparison to other inference algorithms, especially in the performance of the algorithm. It still takes a long time to compute the topic distribution of the data. There are still room for improvement in the time taken for the algorithm to complete the topic distribution. Using two datasets, an evaluation of the performance of the algorithm has been conducted. Results show that Gibbs sampling as the inference algorithm provides a better prediction on the optimal number of topic of the data in comparison to Variational Expectation Maximization (VEM). © 2018, Springer International Publishing AG.