It is a challenging task to identify sentiment polarity in Arabic journals comments. Algerian daily newspapers interest more and more people in Algeria, and due to this fact they interact with it by comments they post on articles in their websites. In this paper we propose our approach to classify Arabic comments from Algerian Newspapers into positive and negative classes. Publicly available Arabic datasets are very rare on the Web, which make it very hard to carring out studies in Arabic sentiment analysis. To reduce this gap we have created SIAAC (Sentiment polarity Identification on Arabic Algerian newspaper Comments) a corpus dedicated for this work. Comments are collected from website of well-known Algerian newspaper Echorouk. For experiments two well known supervised learning classifiers Support Vector Machines (SVM) and Naïve Bayes (NB) were used, with a set of different parameters for each one. Recall, Precision and F_measure are computed for each classifier. Best results are obtained in term of precision in both SVM and NB, also the use of bigram increase the results in the two models. Compared with OCA, a well know corpus for Arabic, SIAAC give a competitive results. Obtained results encourage us to continue with others Algerian newspaper to generalize our model.