A comprehensive framework was developed to detect political bias in Arabic news articles, with a case study focusing on media reporting of the Palestinian issue. The methodology integrates MARBERT contextual embeddings with classical and deep learning classifiers, including SVM, Logistic Regression, Random Forest, and LSTM. The scalability of data processing was ensured through Apache Spark for potential real-time deployment. Experimental results showed that fine-tuned MARBERT embeddings combined with LSTM achieved the highest classification accuracy of 0.87, along with notable improvements in F1-scores across the pro, against, and neutral categories. These findings highlight the effectiveness of domain-specific fine-tuning of transformer models for political bias classification. The study also addressed class imbalance using SMOTE and class weighting strategies, and assessed feature robustness using multiple vectorization techniques.

This paper describes our system submitted to the OSACT7 AdabEval shared task on Arabic politeness detection (TaskA). The task requires classifying Arabic texts into three categories: Polite, Impolite, and Neutral. We systematically explore multiple approaches, progressing from classical machine learning baselines using pre-trained embeddings to fine-tuned transformer models. Our best system leverages MARBERT, a transformer model pre-trained on one billion Arabic tweets, fine-tuned with Focal Loss to handle the significant class imbalance present in the dataset (70% Neutral). We additionally experiment with hybrid approaches combining fine-tuned embeddings with gradient-boosted classifiers and ensemble methods. Our best single model achieves a macro F1 score of 0.84 and an accuracy of 0.90 on the validation set, substantially outperforming classical ML baselines (F1 = 0.42).

Detecting harmful prompts directed at large language models is a critical safety challenge, especially in Arabic where dialectal variation, code-switching, and culturally specific expressions of harm complicate classification. The ArGuard 2026 shared task addresses this problem through two subtasks: binary classification of prompts as safe or unsafe (B1) and fine-grained classification of unsafe prompts into seven harm domains (B2). We participate with a multi-task fine-tuning of MARBERT that shares encoder representations across a binary head and a seven-class granular head. Training combines focal loss with class weights and label smoothing, a hierarchical consistency regularizer that discourages contradictory safe/unsafe and granular predictions, a weighted sampler, and light character-level augmentation. On the official development set the system reaches a binary macro-F1 of 0.9618 and a granular macro-F1 of 1.0000; on the blind test set it obtains binary macro-F1 0.688 and granular macro-F1 0.752. We analyse the substantial development–test gap and discuss limitations of the submitted system.

Detecting hateful content in Arabic memes is challenging due to dialectal variation, sarcasm, cultural references, and the interplay between visual and textual signals. We participate in ArGuard 2026 Track A (Multimodal Hateful Meme Detection) with a multimodal architecture that fuses AraBERT textual representations, ResNet50 visual features, and a lightweight knowledge-graph (KG) ontology vector derived from domain-specific Arabicbinary keywords. The model jointly optimizes (Hateful / Not Hateful) and multi-label finegrained classification through a gated fusion layer with strong regularization (dropout 0.45, weight decay 0.05, label smoothing 0.1). Gradual unfreezing of AraBERT layers, weighted sampling, and Arabic-specific normalization further improve robustness. On the official development set the system reaches a binary macro-F1 of 0.710; on the blind test set it obtains binary macro-F1 0.717. We analyse the contribution of each modality and discuss remaining challenges of fine-grained multi-label prediction under severe class imbalance.

The ANLP_UniSo team submitted a system to Subtask 1 (Named Entity Recognition) of KnowledgeGraphEval 2026, which focuses on nested NER in Arabic and evaluates crossdomain and cross-dialect generalization from the Wojood corpus to the multi-domain Konooz benchmark. The system is a combination of an Efficient GlobalPointer decoder with rotary position embeddings and a MARBERT backbone. To reduce the domain gap, first domainadaptive pretraining (DAPT) is performed on unlabeled Konooz text, and then a pseudolabeling self-training stage is applied. A second model based on AraBERT is trained and combined to the other model using a stacking ensemble to improve prediction diversity. As for the Wojood test set, the best ensemble achieves a micro F1 of 0.9288, increasing to 0.9292 when entity-type-specific decision thresholds are used, compared with 0.9211 for the single MARBERT model. On the official blind Konooz test set, however, the submitted system obtains a micro F1 of 0.7106 (macro F1 = 0.6101, weighted F1 = 0.7013, P = 0.8143, R = 0.6304, accuracy = 0.9948). The lower recall suggests that the decoding threshold chosen based on the Wojood validation data was too strict for the domain-shifted Konooz data. the system study the difference and propose a threshold-recalibration strategy to improve the performance under the target-domain distribution.