NoSpin: Noise and Silence

A Transparent System for Narrative Analysis in News Media

NoSPin is a full-stack narrative analysis system that examines how real world events are framed across print and digital news media. Instead of reducing coverage to simple "Left vs. Right" labels, the system exposes how narratives are constructed through language, sourcing, emphasis, and omission.

Inspiration

Public trust in news media is at a historic low. According to Pew Research Center, only about 56 percent of U.S. adults say they trust national news organizations, meaning roughly 44 percent do not.

At the same time, news consumption has increased dramatically, creating an environment of information overload rather than understanding.

We were inspired by a simple question: If people do not trust the news, how can we help them see how narratives are constructed instead of telling them what to believe?

Noise and Silence was built to make media framing visible, comparable, and auditable.

What It Does

NoSPin is a narrative comparison system that analyzes how major events are covered across diverse news outlets. For any searched topic such as "Iran Crisis" or "Farmers Protest", the system:

Aggregates Coverage: Pulls articles from 25 plus news sources using RSS feeds.
Groups Sources: Automatically sorts outlets by left leaning, right leaning, and center affiliations.
Classifies Narrative: Categorizes articles into Supportive, Critical, or Neutral coverage.
Scores Bias: Assigns each article a transparent Biasness Scale score from 1 to 10.
Synthesizes Comparisons: Generates a summary highlighting shared facts, points of disagreement, and emphasis differences.

The goal is not persuasion. The goal is comparative understanding. The system does not rank truth. It exposes framing.

How We Built It

Frontend (Client)

React Single Page Application: Built with a component driven architecture.
Interactive UI: Features a search bar with trending topics and article cards grouped by narrative stance.
Accessible Design: We replaced infinite scroll with arrow based navigation to reduce cognitive load. The visual encoding is colorblind safe and the layout is compatible with text to speech tools.

Backend (Server)

Python 3 (Serverless): Modular, service oriented design running on Netlify Functions.
Data Ingestion: RSS based aggregation with static source configuration to ensure reliability.
Reliability: The system is stateless and fully mockable, ensuring deterministic outputs for repeatable hackathon demos.

NLP and Analysis Stack

We intentionally avoided opaque machine learning models. Instead, we used rule based linguistic heuristics grounded in journalism research.

Libraries Used:

spaCy: Used for tokenization, sentence segmentation, part of speech tagging, and named entity recognition.

Design Choice: No embeddings. No fine tuning. No black box classifiers. We prioritize explainability over predictive complexity.

Biasness Score Design

Each article is scored using transparent, interpretable dimensions:

Subjective versus factual language density
Attribution balance and source diversity
Use of absolutist terms
Presence of imperative or mobilizing language

Each component is independently measurable and explainable. The score is designed to be auditable, not predictive.

AI Agent Role

The AI agent is used strictly for synthesis, not for scoring. Through epistemic neutral prompting, the agent is instructed to:

Not introduce new facts
Not infer intent
Not express opinions
Not amplify emotionally charged language

The agent surfaces points of common ground and highlights what supportive versus critical coverage emphasizes. This prevents the system from introducing secondary algorithmic bias.

Accomplishments & Challenges

Challenges We Ran Into

Defining Bias: It was difficult to define bias without collapsing into ideology.
False Neutrality: We had to avoid false neutrality while remaining explainable.
Summarization: Designing summaries that reduce noise without oversimplifying complex topics was a UX challenge.
Accessibility: Ensuring accessibility without sacrificing information density required multiple design iterations.

Accomplishments That We Are Proud Of

A fully explainable bias scoring system without ML.
A narrative comparison model that highlights silence, not just disagreement.
A demo safe architecture with no fragile dependencies.
Accessibility first design including colorblind safe visuals and text to speech support.