Attention Decay
I've been fascinated by how quickly news cycles move. A story dominates headlines for a few days, then vanishes completely. I wanted to visualize this pattern.
The inspiration came from this Axios graph showing topic attention over time. I built Attention Decay to track this myself.
How It Works
Twice daily (8am and 8pm UTC), the app pulls headlines from RSS feeds:
- Reuters
- BBC
- NPR
- AP News
Headlines get fed to GPT-4o which extracts story-level topics - not just entities like "Venezuela" but actual storylines like "Venezuela Opposition Crisis" or "Trump Greenland Purchase Bid". The LLM also maps new headlines to existing topics when they're about the same ongoing story.
Topics are categorized (Politics, World Affairs, Technology, etc.) and you can filter by category to focus on what interests you.
The Visualization
The main view is a ridgeline chart showing each topic's attention over time. You can see stories spike and fade, or occasionally persist for weeks. It's a different way to understand the news - less about what's happening today and more about what captures sustained attention.
Tech Stack
React frontend with D3.js for the visualization, Firebase for hosting and scheduled functions, Firestore for storage. The whole thing runs on Firebase's free tier.