Automated destruction identification from visual data plays a critical role in large-scale documentation, humanitarian analysis, and digital archiving of conflict-related events. Within this context, the Nakba-NLP 2026 Workshop introduced a shared task aimed at training and evaluating a binary image classification model to distinguish between destroyed or damaged infrastructure and intact infrastructure. However, the limited dataset size and the visual variability of real-world scenes make this task particularly challenging. This work presents a Swin Transformer–based framework tailored for destruction image classification. The proposed model employs a hierarchical Swin Transformer backbone for robust feature extraction, followed by a multi-layer perceptron classifier for decision-making. To address the limited data issue, transfer learning and a customized training strategy are applied to adapt the model effectively without full end-to-end retraining. Furthermore, a semi-supervised data expansion approach is utilized to enlarge the training set from 1,400 to 10,000 images, improving model generalization and robustness. Experimental results on the official blind test set demonstrate strong performance, achieving an F1-score of 86.55% and an accuracy of 87.81%, ranking 5th in the shared task.

Due to the widespread usage of social media sites and the enormous number of users who utilize irony implicit words in most of their tweets and posts, it has become necessary to detect sarcasm, which strongly influences understanding and analyzing the crowd’s opinions. Detecting sarcasm is difficult due to the nature of sarcastic tweets, which vary based on the topic, region, the user’s attitude, culture, terminologies, and other criteria. In addition to these difficulties, detecting sarcasm in Arabic has its challenges due to its complexities, such as being morphologically rich, having many different dialects, and having low resources. In this research, we present our submission of (iSarcasmEval) sub-task A of the shared task on SemEval 2022. In Sub-task A; we determine whether the tweets are sarcastic or non-sarcastic. We implemented different approaches based on Transformers. First, we fine-tuned the AraBERT, MARABERT, and AraELECTRA. One of the challenges that faced us was that the data was not balanced. Non-sarcastic data is much more than sarcastic. We used data augmentation techniques to balance the two classes, significantly affecting the performance. The performance F1 score of the three models was 87%, 90%, and 91%, respectively. Then we boosted the three models by developing an ensemble model based on hard voting. The final performance F1 Score was 93%.