I've been running a news clustering pipeline as a side project for a while, now tracking >100k news articles per day. Over 164 days, 11.1M articles from 40,000+ sources were tracked, clustered into events (two or more sources covering the same thing), and then events were linked into situations that last days or weeks.
The title number: 11.1M articles were clustered into 941,096 events - 391,230 of which never got another article after their first day. Only 1.2% get any coverage after day three.
Going one level up, the relation inverts: related events are further clustered into what I call "situations", a timeline of happenings over days/weeks/months. 64% of situations are still developing a week in, 44% after two weeks. Headlines die in a day, but the stories underneath run for weeks.
Some more interesting stats I found in the analysis:
-Donald Trump is by far the most mentioned person in news around the world. On average, he's mentioned in 84 new events every single day. He's reported on more frequently than the next 5 people in the most reported list, combined.
-On a typical day, the biggest event represents about 2.6% of that day's new coverage.
-Although most of the tracked publishers are not from the country, the United States is related to over 19% of news events (meaning either the event happened within the US, or it involves the US in some way).
You can visit the site for all the stats I've been able to produce in the analysis, as well as more details on the methodology for producing them.
The data underneath is queryable over REST and MCP if anyone wants to pull their own cut, the pipeline runs 24/7. https://clstr.news/developers
I've been running a news clustering pipeline as a side project for a while, now tracking >100k news articles per day. Over 164 days, 11.1M articles from 40,000+ sources were tracked, clustered into events (two or more sources covering the same thing), and then events were linked into situations that last days or weeks.
The title number: 11.1M articles were clustered into 941,096 events - 391,230 of which never got another article after their first day. Only 1.2% get any coverage after day three.
Going one level up, the relation inverts: related events are further clustered into what I call "situations", a timeline of happenings over days/weeks/months. 64% of situations are still developing a week in, 44% after two weeks. Headlines die in a day, but the stories underneath run for weeks.
Some more interesting stats I found in the analysis:
-Donald Trump is by far the most mentioned person in news around the world. On average, he's mentioned in 84 new events every single day. He's reported on more frequently than the next 5 people in the most reported list, combined.
-On a typical day, the biggest event represents about 2.6% of that day's new coverage.
-Although most of the tracked publishers are not from the country, the United States is related to over 19% of news events (meaning either the event happened within the US, or it involves the US in some way).
You can visit the site for all the stats I've been able to produce in the analysis, as well as more details on the methodology for producing them.
The data underneath is queryable over REST and MCP if anyone wants to pull their own cut, the pipeline runs 24/7. https://clstr.news/developers