Every story gets its value — before anyone reads it.
Every incoming story is sorted automatically: relevant to markets or not, topics, companies and tickers, news value and sentiment. Sorting, filters and every later step build on that.
What you can do with it
- Read the news stream by news value from 0 to 100: what has substance sits on top, with a one-line reason.
- Filter and search by topic, company, ticker, commodity or crypto, even when the word is not in the headline.
- Have opinion pieces, sport, advertising, event notices and general items sorted out automatically.
- See whether a story is positive, neutral or negative, again with a reason.
- Pick the language model freely. On top of that there is Jev, a decision model with finer tags and a confidence score per answer; it is the default for dpa and can be switched on for RSS.
- Classify a selection by hand in the “New” stage, or let classification run in the background all the time, within a daily budget.
- Have stories about directors’ dealings and earnings recognised and linked to the existing record.
The thinking behind it
A newsroom cannot read everything that comes in. So the first question for every story is: is this one for us, and how much does it matter? Answering it costs a fraction of a cent and decides what the more expensive steps, such as fact-checking and research, get spent on later.
Nothing is classified twice. dpa-AFX stories bring their classification from Jev and do not go through a language model again. And when a backlog of more than 270,000 old stories surfaced, we deliberately marked them as “too old” rather than spend money on stories nobody is going to write about.
Sort first, pay later.
Improvements
dpa stories are not classified twice
Stories from the dpa-AFX archive take their topics, news value, sentiment and companies from their Jev classification. A second pass through a language model is no longer needed.
Directors’ dealings and earnings are recognised
Two new topics. After classification, such a story is linked to the existing record and, if there is one, not rewritten a second time.
The backlog shows what is really open
Stories older than 30 days that were never classified now count as “too old”. Since then the “open” counter shows the real backlog instead of a six-figure legacy pile.
Model of your choice, Jev as a second method
You used to be able to set only the provider; now you can set the model too, and leaving it empty means the provider’s default. Jev joins as well, with finer tags and a calibrated confidence per answer.
Daily budget before every batch
Classification checks the daily budget before each batch and stops before it gets more expensive than planned.
Classification as the first step
Every new story gets topics, exclusion reasons and a news value. The more expensive follow-up steps only run above a certain news value.
Give feedback or ask a question
Missing something, bothered by something, unsure about something? Write it down here — it goes straight to the people working on this feature.
Or write to us right here: