Feature-based Approaches for Ethical News Personalisation

Abstract

In recent years, the automation and optimization of content personalisation has become widespread in online mediums. Such recommendation approaches — recommender systems — create increased value to businesses through optimizing business objectives, while providing more accurate suggestions to users, increasing customer satisfaction. In contrast to usual approaches, when considering the automation of news personalisation, ethical journalistic responsibilities have to be taken into account alongside business and user objectives. This thesis evaluates an industry recommender system of a Finnish Media company and describes a novel interpretable model for improving its performance. These improvements are evaluated theoretically and experimentally, showing the marked increase in performance through leveraging previously unused latent information.