BiasLens: Graph Machine Learning Approach for Media Bias Detection

Bias in political media poses a serious concern in an age revolving around media consumption, as it shapes public perceptions and sows social discourse. Most bias detection methods focus solely on the textual content of an article, often overlooking the critical role of authors and news outlets in passing on their biases to articles. In this paper, we introduce BiasLens, a novel framework that leverages Knowledge Graphs (KGs) and graph embeddings incorporating different influences in bias classification. By constructing a KG that captures relationships among articles, authors, and publishers, we can enable a more context-aware bias detection approach. We used a labeled set of political news to construct a KG, applied structured graph-walking strategies, and extracted article node embeddings to train a classifier model. Our approach outperforms traditional text-based methods, achieving an accuracy of 96%, compared to 84% using a BERT-based classifier. These results highlight the potential that KG-based methods have in bias detection tasks and can lay the groundwork for future advancements in media bias detection.