This study investigates the robustness and generalization of transformer-based models for automatic media bias detection. We explore the behavior of current bias classifiers by analyzing feature attributions and stress-testing with adversarial datasets. The findings reveal a disproportionate focus on rare but strongly connotated words, suggesting a rather superficial understanding of linguistic bias and challenges in contextual interpretation. This problem is further highlighted by inconsistent bias assessment when stress-tested with different entities and minorities. Enhancing automatic media bias detection models is critical to improving inclusivity in media, ensuring balanced and fair representation of diverse perspectives.

Media bias detection poses a complex, multifaceted problem traditionally tackled using single-task models and small in-domain datasets, consequently lacking generalizability. To address this, we introduce MAGPIE, a large-scale multi-task pre-training approach explicitly tailored for media bias detection. To enable large-scale pre-training, we construct Large Bias Mixture (LBM), a compilation of 59 bias-related tasks. MAGPIE outperforms previous approaches in media bias detection on the Bias Annotation By Experts (BABE) dataset, with a relative improvement of 3.3% F1-score. Furthermore, using a RoBERTa encoder, we show that MAGPIE needs only 15% of fine-tuning steps compared to single-task approaches. We provide insight into task learning interference and show that sentiment analysis and emotion detection help learning of all other tasks, and scaling the number of tasks leads to the best results. MAGPIE confirms that MTL is a promising approach for addressing media bias detection, enhancing the accuracy and efficiency of existing models. Furthermore, LBM is the first available resource collection focused on media bias MTL.