Exploring Media Bias in News Recommender Systems: From Detection to Mitigation

This thesis explores the issue of media bias in digital platforms, particularly focusing on its detection, its joint impact with news recommendation algorithms on users, and methods for mitigating its effect. The proliferation of digital news platforms like Google News, MSN News, and Flipboard has significantly increased the volume of information available, changing how people access news and presenting opportunities and challenges in news consumption. While beneficial, the ease of assessing various news sources raises concerns about the quality of information, particularly regarding media bias---situations where articles present one-sided viewpoints, either ignoring or attacking opposing views. Media bias can distort perceptions of news, undermining informed public discourse and democratic processes. With the vast volume of news articles, identifying media bias is challenging, making automated media bias detection crucial. However, the scarcity of high-quality, human-labelled datasets poses a significant challenge to developing accurate and efficient bias detection models. This thesis addresses this gap through semi-supervised machine learning methods, leveraging distantly labelled data to improve detection models. Moreover, as news consumers increasingly rely on online platforms, personalized news recommender systems have become influential in shaping news consumption. These systems, while designed to cater to user preferences, may inadvertently exacerbate media bias by creating information bubbles and promoting content similar to users' past choices. This thesis investigates the role of news recommendation algorithms in the proliferation of media bias, examining their sensitivity to biased content in users' reading histories and their long-term effects on news consumption patterns. To mitigate media bias in news articles and recommender systems, this research explores automatic sentence rewriting methods, to improve objectivity in news reporting and reduce biased media proliferation. By addressing the challenges of automated media bias detection, understanding the influence of recommender systems on media bias, and exploring mitigation strategies, this thesis contributes to improving the quality and fairness of news consumption in the digital age. This comprehensive exploration, including the development of media bias detection methods, analysis of recommender systems' role in bias proliferation, and exploration of debiasing strategies, aims to enhance the transparency and credibility of news in the digital era, supporting a more informed public discourse.