Conflicts, Villains, Resolutions: Towards models of Narrative Media Framing

TL;DR

The paper reframes media framing as a narrative-level, multi-label phenomenon by integrating five frames (Resolution, Conflict, Human Interest, Moral, Economic) with narrative roles for key entities (Hero, Villain, Victim). It introduces the Narrative Frames Corpus of 428 English climate-change articles annotated with frames, entity roles, and stakeholder categories, using binary indicator questionnaires and a factor-analysis-based labeling pipeline. A novel Retrieval-Based Frame Prediction (RBF) method combines frame descriptions with sentence-level retrieval and Longformer classification to yield transparent, evidence-backed predictions, achieving strong multi-label performance over baselines. The study reveals systematic differences in framing across political leanings, highlights the interplay between frames and narrative roles, and provides a dataset and methods to scale document-level narrative framing analyses, while outlining avenues for multi-task modeling and cross-language extension. Limitations include annotator agreement variability and the small, English-only corpus, motivating future work on soft-agreement modeling and broader applicability.

Abstract

Despite increasing interest in the automatic detection of media frames in NLP, the problem is typically simplified as single-label classification and adopts a topic-like view on frames, evading modelling the broader document-level narrative. In this work, we revisit a widely used conceptualization of framing from the communication sciences which explicitly captures elements of narratives, including conflict and its resolution, and integrate it with the narrative framing of key entities in the story as heroes, victims or villains. We adapt an effective annotation paradigm that breaks a complex annotation task into a series of simpler binary questions, and present an annotated data set of English news articles, and a case study on the framing of climate change in articles from news outlets across the political spectrum. Finally, we explore automatic multi-label prediction of our frames with supervised and semi-supervised approaches, and present a novel retrieval-based method which is both effective and transparent in its predictions. We conclude with a discussion of opportunities and challenges for future work on document-level models of narrative framing.

Figures (5)

• Figure 1: Association of roles with different frames (top) and stakeholder groups (bottom).
• Figure 2: Distribution of frames across political leanings of news outlets (as attested by MBFC).
• Figure 3: Narrative roles assigned to stakeholder categories in news outlets with different political leaning.
• Figure 4: Our annotation interface. Left: Excerpt of the annotation form which covers the 22 binary indicator questions, and free-text fields to record role-specific entities. Right: the news article with prevalent entities highlighted (based on automatic entity recognition and co-reference resolution.)
• Figure 5: Distribution of the most frequent frame labels.