Mixup, as a technique for augmenting data within the feature space, operates by applying linear interpolation to input instances and their associated modeling targets derived from randomly selected samples. The efficacy of this method in substantially enhancing the predictive accuracy of cutting-edge networks has been established across both image and text classification tasks. Despite its demonstrated success in various contexts, its application within the context of the Arabic language remains an unexplored area of research. This study employed three strategies to adapt Mixup for application in Arabic sentiment analysis. Experimental evaluations were conducted to assess the effectiveness of these strategies, utilizing a range of benchmark datasets. Our studies demonstrate that these interpolation strategies effectively function as domain-independent methods for augmenting data, in the context of text classification. Furthermore, these strategies have the potential to lead to enhancements in performance for both convolutional neural network (CNN) and long short-term memory (LSTM) models.