Grouped, not summarised. Headlines about the same event are
clustered by shared vocabulary, then ranked by how many independent
newsrooms ran them. That is the closest thing here to an objective measure
of what matters. Nothing is rewritten or generated; every line is a real headline,
linked to its source.
How to read this.
Outlets counts distinct publishers, not feeds. BBC and BBC
World are one newsroom, and counting them twice would inflate everything the
BBC covers. Split is the spread of sentiment across the outlets
covering a story: a high number means the same event is being framed very
differently depending on who you read, which is something no single news site
will show you. Clustering is deterministic word-overlap, so it will sometimes
group two stories that share unusual vocabulary, or split one story that
doesn't. Expand a story and you can see exactly what it grouped.
About the sources. Every story here is a headline published
in an outlet's own public RSS feed, shown as written and linked to the
original. No article text, images or generated summaries are copied or
stored. Sentiment scores are ours, computed automatically from the wording of
the headline alone; they are not the publisher's view and not a judgement
about the outlet. Feeds are read once an hour by a bot that identifies itself.
If you publish one of these feeds and would rather not appear here, open an
issue at github.com/epiccoding/Headlines and you'll be
removed.