Abdessalem Kelaiaia1 and Hayet Merouani2

Abstract: Recently, probabilistic topic models such as Latent Dirichlet Allocation (LDA) have been widely used for applications in many text mining tasks such as retrieval, summarization and clustering on different languages. In this paper, we present a first comparative study between LDA and K-means, two well-known methods respectively in topics identification and clustering applied on arabic texts. Our aim is to compare the influence of morpho-syntactic characteristics of Arabic language on performance of first method compared to the second one. In order to, study different aspects of those methods the study is conducted on four benchmark document collections in which the quality of clustering was measured by the use of four well-known evaluation measures, Rand index, Jaccard index, F-measure and Entropy. The results consistently show that LDA perform best results more than K-means in most cases.

Keywords: Clustering, topics identification, arabic text, LDA, k-means, preprocessing.

Received November 7, 2012; accepted November 27, 2013