A two-year research project between the largest newspaper in France and publishing software provider Twipe has been using AI to identify evergreen archive content for republication. The sheer scale of the Ouest-France archive made automated selection the only practical way to identify and select relevant archive articles for republishing at a local and regional level.
Takeaways
- During the project, more than 30 million archive articles were tagged with a content monetization predictive score. This combined metric indicates both the historic value of the article and its relevance within the current news cycle.
- Over 500 articles have been republished, some selected by a journalist and others selected by the algorithm. Results show republished articles selected to be positioned within the news content of the day are generating strong reader engagement.
- The project team has developed an internal search engine based on the content monetization predictive score. This allows local newsrooms to select and re-publish evergreen content with just a couple of clicks.
AI improves efficiency 5X
- Researchers asked journalists to select articles for republication from a random set of 1000 archive articles. On average, 8.5% of articles were selected for republication. A parallel crowdsourcing exercise using 100 readers produced similar results.
- From a pool of archive articles selected by the algorithm, 46% were selected by journalists for republication. This means the algorithms were able to surface the right articles almost five times faster than journalists working independently.
- This improved efficiency was seen as a real asset within Ouest-France local newsrooms, giving staff ‘low effort content’ on slow news days. It also inspired journalists to revisit past topics or add current context to historic coverage.
In some instances republished articles performed better than newly published content. This was particularly true at local level – the best performing articles were accompanied by a very strong image and were related to important local events or personalities.
Not all articles suggested by the algorithm were suitable for re-publishing, however. Some referenced people who had become controversial in the local community. Other articles were related to timely coverage of sporting events that was no longer relevant. Others were largely suitable, but had no associated images.
The future for AI in publishing
AI is increasingly being used in publishing to automate a range of tasks, from surfacing relevant archive content, reporting and personalisation.
- JAMES is an AI ‘digital butler’ that personalises content for readers of The Times and The Sunday Times.
- AI streamlines workflows at the BBC to enable journalists to focus on reporting.
- Facebook is using AI to detect word patterns that may indicate fake news stories.
A 2019 survey of journalists working with AI at 71 news organisations in 32 countries reported that just over a third of respondents to the survey had an active AI strategy. Around 50% of respondents said they used AI for newsgathering and a further two-thirds said they used it for production and just over half for distribution.
Since then, the changes to working practices and the financial pressures brought by COVID-19 threaten media organizations around the world. As the industry tries to recover, it’s a safe bet that the efficiencies brought by AI will see its application become ever more common.
