This paper presents a method for detecting sarcasm in Arabic text using deep learning. It introduces a dataset combining several Arabic datasets annotated for sarcasm and dialect. The methodology employs advanced neural networks like AraBERT to address Arabic linguistic complexities, including dialects and syntax. Central to the approach is an attention interaction module that optimizes sarcasm detection features, enhancing the model’s ability to interpret explicit language and sarcastic tones. Experimental results indicate significant improvements in sarcasm detection accuracy and depth for Arabic texts, potentially setting new benchmarks and offering valuable insights for various applications.