Significance Heatmap

The significance heatmap visualizes news coverage from the last 24 hours, based on article volume and average significance scores. It is rendered as an interactive Plotly treemap where you can click into topic categories to see the source-level breakdown.

Key Metrics

• Size of each block: Proportional to the number of articles in that topic.
• Color gradient: From light blue (low significance, ~0-3) through amber (~5) to dark red (high significance, ~7-10).
• Drill-down: Click a topic block to see which publications contribute articles within it.
• Data source: Aggregated from RSS feeds and web searches across Estonian and English news sources.

7-Factor Scoring Methodology

Each article is scored by an LLM (xAI Grok) on seven dimensions, combined into a single 0-10 significance score using a weighted formula:

Factor	Weight	Description	Example
Scale	4/20	How broadly does the event affect people?	Local crime = 2, Global pandemic = 9
Impact	4/20	How strong is the immediate, tangible effect?	Minor policy tweak = 2, Market crash = 9
Novelty	3/20	How unique and unexpected is this event?	Quarterly earnings = 2, First AI-written law = 9
Potential	3/20	How likely is this to shape the future?	Celebrity gossip = 1, Climate treaty = 8
Legacy	3/20	How likely to be remembered as a turning point?	Sports result = 1, Moon landing = 10
Positivity	1/20	Counteracts negativity bias in news.	War = 0, Breakthrough cure = 9
Credibility	2/20	How trustworthy is the source?	Anonymous blog = 2, Reuters with data = 9

Formula: significance = (scale*4 + impact*4 + novelty*3 + potential*3 + legacy*3 + positivity*1 + credibility*2) / 20

Why Positivity?

News sources have a well-documented negativity bias: they overreport negative events and underreport positive ones. This factor (weight 1/20) brings the ratio closer to 50:50 in the high-significance range. Without it, scores 5+ would mostly consist of wars and disasters. With it, scientific discoveries and tech advancements also surface.

Expected Distribution

Score	%	Typical Content	Example
0-2	~60%	Sports results, entertainment, local news	League match recap, celebrity birthday
3-4	~25%	Regional politics, business earnings	Company quarterly report, city council vote
5-6	~10%	Significant national events, major policy shifts	Central bank rate change, major trial verdict
7-8	~4%	Major world events, breakthroughs	Naval blockade, AI regulation passed
9-10	~1%	Once-in-a-decade events, paradigm shifts	Start of major war, first contact

On a typical day, only 5-15 articles score above 5. If nothing significant happens, the top of the feed is intentionally short.

Philosophy

Significance is objective — it measures how much an event affects humanity as a whole. This is different from importance, which is subjective and personal. News about a family member's health is important to you but not significant to the world. A naval blockade in the Strait of Hormuz is significant to everyone, whether they know it or not.

The goal is to cut through sensationalist noise and surface only what truly matters. Every news site has some notion of "top stories", but they optimize for clicks. We optimize for significance.

Journalist Map

The journalist map visualizes reporter activity over the last 24 hours as an interactive treemap, structured as Publication -> Journalist.

How to Read the Map

• Block size: Proportional to the number of articles published by that journalist in the last 24 hours. A larger block means more articles.
• Color: Represents activity level — cooler blues for lower output, warmer ambers and reds for highly active journalists.
• Hierarchy: Top level shows publications (ERR, Postimees, Bloomberg, BBC News, etc.). Click to drill into individual journalists within each publication.
• Drill-down: Click any journalist name in the list below the map to trigger a chat query that fetches their recent articles with source, significance score, and sentiment rating.

Data Extraction

Journalist names are automatically extracted from RSS feeds and article HTML metadata. The extraction pipeline includes several cleaning steps:

• Role suffix removal: Strips Estonian and English role suffixes like "ajakirjanik" (journalist), "fotograaf" (photographer), "toimetaja" (editor), "correspondent", "reporter".
• Non-person filtering: Removes entries that are URLs, email addresses, organization names (e.g., "ERR", "BBC News", "Postimees"), pipe-separated metadata strings (e.g., "uudised | ERR"), and single-word entries.
• Deduplication: Normalizes names and removes duplicates within the same source (e.g., "Aet Rebane" and "Aet Rebane ajakirjanik" are merged).
• Minimum threshold: Only journalists with 2+ articles in the display window are shown on the map.

Example: Reading a Journalist Entry

Limitations

• Some RSS feeds only provide section names (e.g., "sport | ERR") instead of real journalist names — these are filtered out.
• Paywalled sources (WSJ, FT) may have limited author metadata in their RSS feeds.
• The map refreshes with each page load; it does not auto-update in real-time.

Data Sources

FastMedia aggregates from 7 news sources across 2 languages:

Source	Language	Type
ERR	Estonian	Public broadcaster
Postimees	Estonian	Daily newspaper
Eesti Paevaleht	Estonian	Daily newspaper
BBC News	English	Public broadcaster
Bloomberg	English	Financial news
Financial Times	English	Financial newspaper
Wall Street Journal	English	Financial newspaper

Articles are fetched via RSS every 5 minutes, with full text extracted using newspaper4k. Additional articles are discovered via Tavily and Exa web search APIs.

Inspired by News Minimalist. Powered by xAI Grok.