Advancing Media Objectivity: A Deep Learning Model for Detection of Bias in Amharic News

Detecting bias in news stories is critical for maintaining media objectivity and fostering a well-informed public. This task becomes particularly challenging for languages with limited natural language processing resources, such as Amharic. We propose a novel method using convolutional neural networks, a deep learning technique, to identify bias in Amharic news articles. To address this problem, we compiled a diverse dataset of Amharic news stories from social media, encompassing various topics and viewpoints. The domain experts meticulously annotated the dataset, classifying each article as neutral or biased based on the underlying content. Our model, trained and evaluated on this annotated dataset, effectively extracts linguistic patterns and contextual information from Amharic text. Through hyperparameter optimization, we achieved high accuracy (89.5%) and precision (fl-score of 80.28%) in bias detection on a separate test set. This research contributes to improved media literacy, promotes objective reporting, and empowers the Amharic-speaking community to become more discerning consumers of information.