Methodology

How News Lens works. No black boxes. Full transparency.

1. Our Philosophy

News Lens exists to make news framing visible. We believe that understanding how a story is told is as important as understanding what happened.

We deliberately avoid making judgments about accuracy, bias, or trustworthiness. These are complex assessments that require context we don't have. Instead, we focus on showing you the raw material: the same event, covered by different sources, with different framings.

2. Source Selection

We track 33 news sources across 5 regions. Our selection criteria:

• Reach: Major outlets with significant readership
• English availability: Must have English-language content
• RSS accessibility: Must provide machine-readable feeds
• Editorial diversity: Include different perspectives within each region

Each source is labeled with its editorial positioning: a brief, factual description of the outlet's ownership, funding model, and general editorial stance. These labels are not judgments; they're context.

Sources by Region

United States 6 sources

Europe 7 sources

Full source list with RSS feeds and positioning notes available in our public data repository.

3. Story Clustering

We group articles about the same event using semantic similarity of headlines. Here's how it works:

1. Ingestion: We poll RSS feeds every 15 minutes
2. Embedding: Each headline is converted to a vector representation
3. Matching: New articles are compared against recent clusters
4. Clustering: Articles above a similarity threshold are grouped together

Parameters

Parameter	Value	Explanation
similarity_threshold	0.65	Minimum cosine similarity to join a cluster
time_window	48 hours	Clusters older than this are harder to match
decay_factor	0.1	Similarity penalty per day for old clusters

4. Framing Tags

We detect 10 mechanical framing signals in each article. These are objective, detectable patterns, not subjective judgments about quality or bias.

Tag	What We Detect
Quotes Government	Article cites government officials, ministers, or spokespersons
Quotes Opposition	Article includes critics, opposition leaders, or dissenting voices
Mentions Casualties	Article reports deaths, injuries, or casualties
Cites Data	Article includes specific numbers, percentages, or statistics
Cites Document	Article references a specific report, filing, or primary source
Historical Context	Article provides background or timeline information
Expert Quote	Article quotes academics, analysts, or researchers
Future Speculation	Article predicts or speculates about future outcomes
International Reaction	Article includes responses from other countries
Economic Impact	Article discusses financial or economic consequences

Detection Method

Tags are detected using a combination of:

• Keyword matching: Presence of specific terms (e.g., "minister said", "percent")
• Pattern matching: Regex patterns for common phrasings
• Entity recognition: Identification of quoted sources

Each tag includes a confidence score (0-1) based on the strength of the match. We only display tags with confidence above 0.3.

5. Limitations

News Lens has significant limitations you should understand:

• English only: We only process English-language content, missing coverage in local languages
• RSS dependency: We can only track sources with accessible RSS feeds
• Headline-based clustering: We may miss articles with unusual headlines or group unrelated articles with similar headlines
• No full-text analysis: Framing tags are detected from headlines and descriptions, not full articles
• No verification: We don't verify facts or check accuracy
• Selection bias: Our source list reflects choices about what's "major" and "representative"

6. Updates & Feedback

This methodology is a living document. We update it when we change our approach.

Last updated: January 2026
Version: 1.0

Found an error? Have a suggestion? Email us at [email protected]