ANTI: a framework for agonistic news recommendation

Abstract

Algorithmic news recommendation on news aggregator platforms increasingly shapes what news perspectives people encounter. However, in an effort to increase diversity and reduce the personalized ‘filter bubble’ effect, many conflicting viewpoints are often blended into a single feed. This thesis explores how an agonistic approach to news recommendation might better support users in interpreting diverse and oppositional perspectives. We introduce the Agonistic News Topic Interpreter (ANTI), a design concept developed through a Research through Design methodology. ANTI draws on theories of framing and agonistic pluralism to structure news diversity more intentionally: through the use of frame personae. Two practical design strategies are presented: persona profiles, that contextualize the personae and provide representative articles, and substitution articles, intended to recommend users alternative perspectives on similar topics. Using the MIND and Media Frame Corpus datasets, we train a frame classifier and model frame personae for three cases: abortion, immigration, and the Hong Kong protests. The results show that the personae effectively separate perspectives and that these are nicely contextualized by the persona profiles, enabling more interpretable diversity. Substitution articles show potential but require further refinement to match events across articles. We conclude this thesis by summarizing defining elements of an ANTI and agonistic news recommendation, outlining (agonistic) design guidelines for current news platforms and suggesting future research opportunities.