Talking Point based Ideological Discourse Analysis in News Events

TL;DR

This work tackles the challenge of extracting and comparing ideological discourse in news events by introducing a structured talking-point framework that represents articles as relational talking points and clusters them into Prominent Talking Points ($PTP$s). An LLM-based pipeline extracts talking points, forms $PTP$s via clustering, and generates ideology-specific partisan perspectives (left vs. right) for each $PTP$, enabling event-level (via the event $\mathcal{E}$) and $PTP$-level analyses. The authors release a dataset of 6,141 articles across 24 events from 126 bias-coded outlets, along with $PTP$s and partisan perspectives, and validate the approach through ideology and partisan classification tasks, plus human evaluation and narrative event snapshots. Results show that TopK Event Partisan View improves event-level ideology classification and that Partisan Perspectives yield strong PTP-level signals, with Narrative-LLAMA providing further gains; the framework supports robust event-level discourse analysis while acknowledging limitations and ethical considerations for deployment.

Abstract

Analyzing ideological discourse even in the age of LLMs remains a challenge, as these models often struggle to capture the key elements that shape real-world narratives. Specifically, LLMs fail to focus on characteristic elements driving dominant discourses and lack the ability to integrate contextual information required for understanding abstract ideological views. To address these limitations, we propose a framework motivated by the theory of ideological discourse analysis to analyze news articles related to real-world events. Our framework represents the news articles using a relational structure - talking points, which captures the interaction between entities, their roles, and media frames along with a topic of discussion. It then constructs a vocabulary of repeating themes - prominent talking points, that are used to generate ideology-specific viewpoints (or partisan perspectives). We evaluate our framework's ability to generate these perspectives through automated tasks - ideology and partisan classification tasks, supplemented by human validation. Additionally, we demonstrate straightforward applicability of our framework in creating event snapshots, a visual way of interpreting event discourse. We release resulting dataset and model to the community to support further research.

Figures (11)

• Figure 1: Shows how each political orientation discusses an event about Climate Change. (a) Shows a collection of talking points extracted from all the news articles about the event. (b) Shows an example of a Prominent Talking Point (PTP), a repeating theme discussed by both political ideologies. (c) Shows left- and right- ideological interpretations of the PTP.
• Figure 2: Framework overview: Event-specific news articles are organized into a relational structure, referred to as talking points. These are clustered to identify Prominent Talking Points (PTPs), a vocabulary of repeating themes that are relevant to the event. Each PTP is infused with ideological information to identify left-leaning and right-leaning viewpoints, referred to as partisan perspectives.
• Figure 3: Shows Aggregated metadata corresponding to the PTP shown in Figure \ref{['fig:partisan_perspective']}. Each target is paired with its most common actor, sentiment, and media frame within the actor-target context. Metadata is color-coded: Blue for left ideology and Red for right ideology.
• Figure 4: Discourse snapshot for a Covid-related event. Numbers represent PTP IDs. Points near center denote bipartisan discussion.
• Figure 5: Shows an example of the generated perspectives for a right-leaning news article related to Biden's executive orders. The generated perspectives from Narrative-LLAMA capture broader ideological discourse as opposed to its counterparts.