Automated identification of bias inducing words in news articles using linguistic and context-oriented features

• DOI: 10.1016/j.ipm.2021.102505
• Publication Date: 2021-02-11T22:02:11Z
• Authors: Timo Spinde; Lada Rudnitckaia; Jelena Mitrović; Felix Hamborg; Michael Granitzer; Bela Gipp; Karsten Donnay
• Keywords: Identification; Media Bias

Media has a substantial impact on public perception of events, and, accordingly, the way media presents events can potentially alter beliefs and views public. One ways in which bias news articles be introduced is by altering word choice. Such form very challenging to identify automatically due high context-dependence lack large-scale gold-standard data set. In this paper, we present prototypical yet robust diverse set for research. It consists 1,700 statements representing various instances contains labels identification sentence level. contrast existing research, our incorporate background information participants' demographics, political ideology, their opinion about general. Based data, also detect bias-inducing words automatically. Our approach feature-oriented, provides strong descriptive explanatory power compared deep learning techniques. We engineer linguistic, lexical, syntactic features that indicators. resource collection most complete within research area best knowledge. evaluate all combinations retrieve possible importance both future task Machine Learning approaches with features. XGBoost, decision tree implementation, yields results. achieves an F1-score 0.43, precision 0.29, recall 0.77, ROC AUC 0.79, outperforms current detection methods based propose improvements, discuss perspectives feature-based combination neural networks system.