Focus on the few-shot approach has increased recently for TC as it is competitive with fine-tuning models that need a large dataset [14]. In NLP, the process of using PTMs to classify new data is preferable to the expensive process of training a model from scratch. This can be considered a kind of TL, i.e., it focuses on reusing knowledge of PTMs to solve different problems, as long as the pre-training data is appropriately comparable. Transferring knowledge allows the model to circumvent the lack of data and enable FSL as a low-cost solution. To clarify, the term shot refers to a single example that is used for training, and the number of examples available for training is equal to N in N-shot learning. The focus of this study is on few-shot classification, which involves distinguishing between N classes using K examples of each. In this approach, N-way-K shot classification implies that each task involves N classes with K examples. In FSL, the model is able to predict a new class based on a few new examples [11] by transferring knowledge and contrasting examples. Such contrastive learning [5] has shown its effectiveness in different studies of various NLP tasks [20]. However, as far as we know, no previous studies have applied contrastive learning to standard Arabic for multi-class classification. This study aims to apply few-shot learning using a Siamese Network-based model(SN-XLM-RoBERTa [6]) to classify MSA texts in predefined classes labelled with the most common ministriesβ names. For this study, we extracted a new dataset from an AI-powered research tool. The model was fine-tuned by K examples per class. We experimented with various K values, including 10, 20, 50, 100, and 200. The results show that the accuracy in distinguishing between 6 classes using 200 examples of each is 91.076%. Moreover, the results indicated that employing few-shot learning, as in SN-XLM-RoBERTa, in classifying MSA texts can be a promising solution in case of an insufficient dataset or uncertain labelling. Few-Shot Learning (FSL) may contribute to the research domain by automating the classification process.