How Siren agency creates 16,000 news articles about prosecuted crimes in Sweden
Reading Time 3 mins | April 19, 2021 | Written by: AX Semantics
Siren, a news agency based in Sweden, delivers more than 3,000 news articles a day to the country's publishers - customized to each user's needs. The news agency uses AX Semantics` natural language generation software to create messages based on law enforcement, criminal court orders and district court judgments. In addition, background information such as corporate holdings and prior criminal offenses is added.
In this interview, Matti Larsson, CEO of Siren News Agency shares his experience about the software and describes how Siren creates simple and meaningful content in which crimes, defendants, locations and sentences appear.
What do you use the automated text generation software for?
We collect all sentences and prosecution documents in Sweden and generate written descriptions from them.The software makes its own judgement on news value from a specific number of variables, including some background information on the defenders.
A crime involving a successful businessman involved in a serious crime will have a higher news value than a youngster involved in a petty crime. The software also checks previous convictions.
How many texts have you already generated/are you planning to generate?
We generate about 16.000 news articles every month, covering basically every crime prosecuted in Sweden. It is then up to our customers in the various media outlets to determine which articles go to publication, and if they want to enrich the texts with quotes from the prosecutor. Some of them do, and some don’t.
What is the challenge with AX Semantics software?
The heavy part of this process is to make sure you have extracted all the datapoints from the source material, before sending it to AX Semantics. The data preparation part in transformer view took a lot of time, specifically the first time. But the learning curve was flattened a great deal by having support from AX Semantics team who could work in the same view we had access to and you could see exactly what they were doing.
How many people in which roles are working with AX Semantics?
Only one so far. As we plan new feeds, we might educate more staff on how to work with AX Semantics.
Do you already see effects by using automated texts? If yes, which ones?
Using AX Semantics instead of usual templates allows more varied and richer content. When launching the crime software, we saw an immediate rise in our read logs. The bar chart shows how the number of readers increased in the monthly interval when we launched the feeds. The green and blue bars reflect the number of read articles created with automated content generation. The purple colored bars are the manually created articles.
The pie chart shows the proportions of articles read from the two crime feeds in blue and purple that were generated with automated content generation. The green proportion is the manually created content.
What are your plans for the future (more text types, languages, etc.)?
We have hundreds of feeds that run through our usual processes. We are already checking which feeds would be more beneficial to generate with AX Semantics.
How would you describe the AX Semantics software in one sentence to someone who does not know the tool yet?
The automated content generation with AX Semantics meets our high standards and criteria that we would also have for our manual text creation.