Mapping News Narratives Using LLMs and Narrative-Structured Text Embeddings

TL;DR

This work tackles the challenge of quantifying narratives in large-scale news data by grounding a narrative representation in Greimas' Actantial Model and embedding-based semantics. It introduces a pipeline that uses zero-shot prompting to extract six actants per article, converts them into six embeddings, concatenates them into a narrative-structured embedding, and applies dimensionality reduction and clustering to identify cultural narratives. An Israel–Palestine case study across 5,342 articles from Al Jazeera and The Washington Post reveals 18 narrative clusters and distinct editorial patterns tied to actant structure, demonstrating the method's ability to differentiate narratives beyond topic content alone. The proposed framework offers a generalizable tool for cross-topic narrative analysis, enabling comparisons across sources and time with explicit attention to narrative form as well as content.

Abstract

Given the profound impact of narratives across various societal levels, from personal identities to international politics, it is crucial to understand their distribution and development over time. This is particularly important in online spaces. On the Web, narratives can spread rapidly and intensify societal divides and conflicts. While many qualitative approaches exist, quantifying narratives remains a significant challenge. Computational narrative analysis lacks frameworks that are both comprehensive and generalizable. To address this gap, we introduce a numerical narrative representation grounded in structuralist linguistic theory. Chiefly, Greimas' Actantial Model represents a narrative through a constellation of six functional character roles. These so-called actants are genre-agnostic, making the model highly generalizable. We extract the actants using an open-source LLM and integrate them into a Narrative-Structured Text Embedding that captures both the semantics and narrative structure of a text. We demonstrate the analytical insights of the method on the example of 5000 full-text news articles from Al Jazeera and The Washington Post on the Israel-Palestine conflict. Our method successfully distinguishes articles that cover the same topics but differ in narrative structure.

Figures (13)

• Figure 1: Schema of the embedding process. Each actant is embedded using text embeddings that are then reduced in dimension using SVD. Next, we concatenate the reduced embeddings into the narrative-structured embedding. Lastly, we project the embedding to a plane for visualization and clustering.
• Figure .1: Average actant cosine similarity for whole actant corpus over different levels of dimension reduction. The starting point is E5-Large embedding with dimension $1024$.
• Figure 2: Prompt used to extract the Actantial Model from a news article. During inference {{ article }} is replaced with a full-text news article.
• Figure .2: Cosine similarity of key actors within the dataset using E5-Large embeddings.
• Figure 3: Number of articles per week for Al Jazeera and The Washington Post. With a total number of 5342 articles.