With the vast amount of textual data generated daily on Internet platforms, automatic text summarization emerges as a solution for the quick understanding of public opinions and for providing valuable insights for decision-making. This work introduces a comprehensive framework that combines sentiment classification with multi-document text summarization of Facebook comments written in standard Arabic and Moroccan Arabic. In the first stage of our approach, we compare a lexicon-based method against two machine-learning classifiers, Naïve Bayes and SVM. The classification results showed that SVM outperformed the other classifiers. In the second stage, we apply an extractive method for multi-document summarization through sentence clustering using K-means. The evaluation of the extracted summaries was carried out using the standard metrics ROUGE1 and ROUGE2. The results were promising, as we achieved scores comparable to the related work while taking into account the challenging aspects of standard and dialectal Arabic expressions that are employed on social networks.