Original scientific paper
https://doi.org/10.2498/cit.1001770
Arabic Text Classification Framework Based on Latent Dirichlet Allocation
Mounir Zrigui
; LaTICE Laboratory (Research Unit of Monastir ), University of Monastir, Tunisia
Rami Ayadi
; Faculty of Economics and Management, University of Sfax, Tunisia
Mourad Mars
; Stendhal University, Grenoble, France
Mohsen Maraoui
orcid.org/0000-0001-6598-7465
; University of Monastir, Tunisia
Abstract
In this paper, we present a new algorithm based on the LDA (Latent Dirichlet Allocation) and the Support Vector Machine (SVM) used in the classification of Arabic texts.
Current research usually adopts Vector Space Model to represent documents in Text Classification applications. In this way, document is coded as a vector of words; n-grams. These features cannot indicate semantic or textual content; it results in huge feature space and semantic loss. The proposed model in this work adopts a “topics” sampled by LDA model as text features. It effectively avoids the above problems. We extracted significant themes (topics) of all texts, each theme is described by a particular distribution of descriptors, then each text is represented on the vectors of these topics. Experiments are conducted using an in-house corpus of Arabic texts. Precision, recall and F-measure are used to quantify categorization effectiveness. The results show that the proposed LDA-SVM algorithm is able to achieve high effectiveness for Arabic text classification task (Macro-averaged F1 88.1% and Micro-averaged F1 91.4%).
Keywords
LDA; Arabic; stemming algorithm; text classification; SVM
Hrčak ID:
85083
URI
Publication date:
30.6.2012.
Visits: 3.452 *