Multi-Category Support Vector Machines for Identifying Arabic Topics

Mourad Abbas (1) , Kamel Smaïli (2) , Daoud Berkani (3)

1 CRSTDLA - Centre de Recherche Scientifique et Technique pour le Dévelopement de la Langue Arabe
2 PAROLE - Analysis, perception and recognition of speech
3 Department of Electronics [El Harrach]

It is known that Support Vector Machines were designed for binary classification. Nevertless, it would be fruitful to extend this operation to what is called Multi-category classification. That is why, Multi-category Support Nector Machines (MSVM) become nowadays the current subject of several serious researches, aiming to achieve high levels of multi-category classification tasks. This technique has been assessed recently recently in some fields as text categorization, Cancer classification, etc. We should notify that experiments which have been realized until now using MSVM are limited to small data sets, since its computation is more expensive. In this paper, we are interested in the use of this method, for the first time in topic identification. The experiments conducted concern topic identification of Arabic language. The corpora are extracted from ALWATAN newspaper. Achieved results lead to an improvement of MSVM performance i comparison to the baseline SVM method. Nevertheless, SVM still outperforms MSVM when using larger sizes of the vocabulary.

Mots clés: MSVM; Topic identification; Arabic Language

Domaines: Informatique et langage [cs.CL]

Mourad Abbas, Kamel Smaïli, Daoud Berkani. Multi-Category Support Vector Machines for Identifying Arabic Topics. 10th International Conference on Intelligent Text Processing and Computational Linguistics - CICLing 2009, Mar 2009, Mexico, Mexico. ⟨inria-00403102⟩