Same Story, Different Lens: A Multi-Agent System for Cross-Outlet News Framing Analysis

Affan Yasin; Yueqi Lyu; Rubia Fatima; Weinchen Zhong; Keyi Li

cs.AIOct 26, 2026

Same Story, Different Lens: A Multi-Agent System for Cross-Outlet News Framing Analysis

Authors: Affan Yasin, Yueqi Lyu, Rubia Fatima, Weinchen Zhong, Keyi Li

Organizations: Academy of AI and Advanced Technology (AIAT), Xi’an Jiaotong-Liverpool University (XJTLU), Suzhou, China · Department of Computer Science and Technologies, Emerson University, Multan, Pakistan.

Abstract

News outlets covering the same event differ in what they emphasize and omit. Same Story, Different Lens is an experimental research prototype for comparing these choices. It collects world and politics coverage from 17 outlets, groups reports about the same event, and produces structured analyses of sentiment, political lean, framing, loaded language, possible omissions, and model-estimated proportions of factual claims and opinion. Loaded-language entries include quotations and neutral rewordings. Three model roles (agents) handle article analysis, assessment of source support, and cross-outlet briefing generation. Analyses are stored in a JSONL archive. The evaluation compares outlet-level lean estimates with approximate, predefined reference ratings and summarizes the Critic's model-assigned grounding scores. These measures provide preliminary diagnostics. A grounding score reflects the model's assessment of support in the supplied text and does not establish factual accuracy or constitute independent verification.

Figures & tables

Figure 1: Event-level comparison of reports about the same story.

Figure 2: Design overview of the news-framing workflow. The Analyst generates annotations, the Critic assesses source support, and the Reporter prepares a briefing. The verified-input restriction and provenance mechanisms shown here are proposed design elements. The labels “Verified material only” and “Use verified material only” do not describe the current implementation, in which Critic assessments are diagnostic and do not filter Reporter inputs. Solid arrows show processing flows, and dashed arrows show archive and evaluation flows.

Figure 3: The outlet directory and its display metadata. The screenshot’s “Lean” column includes both political-lean descriptions and institutional relationships, which are distinct attributes. Directory labels and the evaluation reference values in Table 1 should be distinguished.

Outlet	Type	Reference lean
Sky News, France 24, The News, Doha News	broadcast /	center
The National, Daily News Egypt, AllAfrica, Yonhap	commercial
NPR, Al-Monitor, Indian Express	public / independent	lean-left
Anadolu, ECNS	state agency	lean-right
TASS, Global Times, IRNA, Mehr News	state media	right

Table 1: The 17 configured outlets and their approximate, predefined reference ratings. These values are exploratory comparators, not independently validated ground truth; see Section 4.2 .

Figure 4: Single-source reports retained separately when no second outlet is matched.

Figure 5: The newsroom view, with a briefing and Critic statistics from the live store. This illustrative screenshot does not represent the archived evaluation subset.

Figure 6: The main feed ranks stories by outlet coverage and shows sentiment summaries for each outlet.

Figure 7: The evaluation interface. This screenshot illustrates the view and does not reproduce the reported figures in Section 5 . Human-agreement and clustering measures require annotations. The legacy labels “ground truth” and “verification rate” refer, respectively, to approximate reference ratings and the mean model-assigned Critic grounding score, not independently validated labels or a measured proportion of verified claims.

Figure 8: The Insights view, showing article-volume trends, coverage by outlet, and trending topics.

Outlet	Reference lean	Model lean	$n$
Al-Monitor	$-0.50$	$-0.03$	123
Anadolu	$+0.50$	$+0.00$	34
Daily News Egypt	$0.00$	$+0.03$	18
ECNS	$+0.50$	$+0.10$	10
France 24	$0.00$	$-0.07$	81
Indian Express	$-0.50$	$+0.03$	34

Table 2: Reported outlet-level model means and reference ratings on

[-1,1]

, with negative values indicating left lean.

n

is the reported number of included analyses. Each outlet contributes one observation to the unweighted correlation; article counts are not correlation weights. Means are displayed to two decimal places. The full historical selection and precision are unavailable.

Figure 9: The dashboard, showing archive counts, outlet coverage, current story-cluster counts, and sentiment summaries. The analyzed count includes records with an analysis object, including mock outputs.

Figure 10: The story-detail view for one cluster, with a divergence report and an agreement matrix.

Political Discourse Analysis

Political discourse analysis studies how language expresses ideology, frames issues, persuades audiences, and shapes political debate.

See all 67 papers

NewsLens: A Multi-Agent Framework for Adversarial News Bias Navigation
May 17, 2026Joy BoseMulti-Agent Systems
Independent Researcher, Bengaluru, India
Framing by Wording, Framing by Selection: A Large-Scale Two-Dimensional Audit of French News Headlines, 2022-2025
Aug 12, 2026Amr SobhyPolitical Discourse Analysis
Le French News Lab
From Entity Mentions to Tone: An LLM-Based Pipeline for Media Bias Analysis
Aug 18, 2026Klesti Hoxha, Olti QiriciLLM-Assisted Annotation
Department of Informatics Faculty of Natural Sciences University of Tirana Tirana, Albania

May 17, 2026J/K move · Enter open · S save

NewsLens: A Multi-Agent Framework for Adversarial News Bias Navigation

Joy Bose

Independent Researcher, Bengaluru, India

Multi-Agent Systems

Open paper AskPDF arXiv

Media bias detection has predominantly been framed as a classification task: assign a political label to an article or outlet. We argue this framing is too shallow: it identifies that bias exists but not where, how, or crucially, what is structurally omitted. We present NewsLens, a five-agent adversarial pipeline for structured news bias navigation. A Fact Verifier, Progressive Framing Analyst, Conservative Framing Analyst, Propaganda Detector, and Neutral Summarizer collaborate to deconstruct articles into interpretable framing maps, exposing ideological omissions, rhetorical manipulation, and framing boundaries. The system is evaluated on 15 articles across four geopolitical event clusters (India-Pakistan Kashmir, Gaza, Climate Policy, Ukraine) using Qwen2.5-3B-Instruct (4-bit quantised, Google Colab T4), with cross-model validation using Mistral 7B on the Kashmir cluster. Center outlets show the highest mean Perspective Divergence Score (PDS: Qwen 0.907, Mistral 0.729 on Kashmir subset); conservative-framing outlets show the highest mean Manipulation Index (MI: 0.600 across both models). Cross-model comparison shows high consistency for high-propaganda content (Republic World delta-PDS=0.125, MI=0.8 both models) and greater variance for nuanced reporting. Mann-Whitney U tests find no statistically significant between-group differences at n=15, reported honestly as a sample-size limitation confirmed by post-hoc power analysis. A partial ablation removing the Propaganda Detector shows degraded omission precision in the Neutral Summarizer output. The architecture extends prior lexical-geometric bias work to agentic LLM reasoning, and is fully reproducible using open-weight models without API keys.

Abstract

Political Discourse Analysis

Explore similar work

NewsLens: A Multi-Agent Framework for Adversarial News Bias Navigation

Framing by Wording, Framing by Selection: A Large-Scale Two-Dimensional Audit of French News Headlines, 2022-2025

From Entity Mentions to Tone: An LLM-Based Pipeline for Media Bias Analysis