in General, ROI

DPG Media; “Trendskout algorithm is our first line of defense for inappropriate reader responses.”

Readers’ responses to news items are not only food for thought on newspaper websites, but also beyond. Many editors are worried about how they can maintain the balance between sufficient participation and a civilized debate in commentary sections. Where many news media in recent years decided to completely disable responses, DPG took a different approach with strict moderation of responses. Recently, selection is not only done by human hands, but also artificial intelligence is hidden behind the response button. The AI platform from Trendskout immediately filters inappropriate responses. Heiko Desruelle and Kurt Minnen from DPG explain.

250 000 reactions a month

“We publish more than 700 articles on our news website every day. Together they account for around 250,000 readers’ responses per month, “says Kurt Minnen. As managing editor, Minnen is the link between the newspaper editors and the technical teams behind the website. “With our extensive commentary section, HLN distinguishes itself from other news media. In the past, all readers’ responses were checked manually by a team of a few permanent employees, supplemented by temporary employees. We have been working with an external agency for the past three years. ”

Heiko Desruelle (left) en Kurt Minnen (right) from HLN

“Articles that go viral lead to a flood of reactions. Such peaks put a lot of pressure on our moderators.”

Pressure on moderators

“Of course, such active moderation comes with a cost,” explains Heiko Desruelle, digital manager at HLN. “A full-time team, both internally and via outsourcing, is not cheap. In terms of content too, it is not always that simple. All moderators first receive training on our internal selection criteria for responses. Although the rules of conduct are clear, a great deal remains open to interpretation. Hate reports and racist reactions are obviously not acceptable and are immediately rejected, but other comments are much less obvious. Often it just depends on the moderator whether a certain reaction is allowed. Sometimes the job also causes stress. Popular articles soon lead to a flood of reactions. Peaks of this kind put a lot of pressure on moderators. And that is exactly where the artificial intelligence of Trendskout turned out to offer a solution. ”

Main issues

  • Moderation of reader responses is not part of the core business of HLN or News City, the underlying news editor of owner DPG Media.
  • Both full-time internal moderators and outsourcing result in considerable wage costs.
  • Fully manual moderation without tools offers relatively little job satisfaction.
  • Unexpected peaks in reader responses were difficult to handle manually.
  • Assessments of individual moderators are not always 100% concurrent and objective.

AI as the solution

“A solution that could reduce the workload of our team without much hassle or investment was more than welcome,” continues Desruelle. “When the people from Trendskout first came to visit us, they brought some relevant business cases with them. I then challenged them with an old dataset from our website: could the AI platform from Trendskout immediately start using it, without further information? Amazingly enough, that was immediately possible. ”

Minnen adds: “Trendskout turned out to be a relief compared to other AI platforms and expensive consulting firms, which demanded lengthy implementation processes. For content modeling at HLN, we simply wanted a clear, inexpensive solution that could immediately provide us with ROI. Trendskout had a live test case running on our website in less than four weeks. It immediately turned out to be a success. Forty percent of the reader responses are now automatically rejected by the algorithm, while doubtful cases end up nicely with a human moderator. ”

“Trendskout had a live test case on the HLN website in less than four weeks.”

Scaling quickly for maximum ROI

  • Functional proof of concept with real company data in a few days.
  • Trendskout software autonomously put together a suitable model based on text (natural language processing) and sentiment analysis.
  • Follow-up project of three months fully operational within four weeks after the green light from HLN management.
  • No risky technical integrations or platform migration.
  • Assessments of individual moderators are impossibly 100% concurrent and objective.

One AI platform for all projects

“There are plans to use the power of artificial intelligence in other places as well. Desruelle continues: “We already have plans to use Trendskout’s AI platform to help us with another tricky issue: the HLN paywall. At the moment our editors place articles behind a payment wall based on their own experience. This manual selection often turns out to be more wet finger work than we would like. We would like to see if the Trendskout AI can discover the winning formula for the selection of premium items.

“That’s the beauty of the Trendskout platform,” continues Minnen. “We can work creatively ourselves to tackle specific business problems, without fuss or risky investments. That gives us a crucial advantage in today’s competitive media landscape. ”

“The Trendskout AI just does what it promises. That ensures ROI from day one. ”

Results in numbers

40

40% of responses are automatically rejected.

40

40% reduction in workload for human moderators.

3

Follow-up project on the same platform in progress within three months of the introduction of AI moderation.