Methodology

We combine curated RSS feeds with the GDELT 2.1 document API to keep a balanced, real-time view of Indian and global coverage.

Source collection

  • RSS feeds from national, business, tech, and world outlets.
  • GDELT query expansion for breaking headlines and global coverage.
  • Feed health checks ensure broken feeds are skipped, not fatal.

Deduplication & threading

  • URLs are canonicalized to detect duplicates.
  • Title similarity + time proximity groups articles into story threads.
  • Story sentiment aggregates article scores with recency & confidence weighting.

Sentiment scoring

  • Transformer sentiment (when enabled) generates positive/neutral/negative scores.
  • Short or failed transformer inputs fall back to VADER for stability.
  • “Why this score” surfaces top phrases and model confidence.

Credibility badges

Credibility badges are optional and reflect general ratings concepts inspired by AllSides or Ad Fontes. We do not claim their data unless it is explicitly imported.

Limitations

  • Sentiment is a directional signal, not a factual assessment.
  • Coverage depends on feed freshness and GDELT availability.
  • Story clustering can occasionally merge similar but distinct events.