AUTOMATED POLITICAL BIAS CLASSIFICATION IN NEWS AGENCIES: A SPARSITY-INDUCING FEATURE SELECTION APPROACH

Abstract

This study offers a new solution to political bias classification in news agencies. Our method uses search engine score functions to develop a measure of the relevance of each word in text scrapped from news websites. With these scores, we train models using existing feature selection methods and a custom feature selec-tion algorithm that we developed. The result-ing models are contrasted with each other and BERT-based counterparts. Models trained using our proposed method and custom algorithm outperformed others by achieving macro F1 scores of 0.81 and 0.78 on right-wing and left-wing bias detection respectively, which outper-form transformer-based classifiers by over 0.30.