Predicting the success of news

Abstract

Traditional recommendation systems have limited possibilities to optimise business value in editorial decision making in news production, as they select the recommendations only from the content whose production has been decided editorially in the daily news process or content from existing content inventories. This paper explores an approach to use predictive analytics to make it possible to optimise story assignment and editing in daily editorial work based on selected business objectives already before publishing. In this case study exploration, we use the 'constructive approach' as a method to provide solutions to concrete business problems with a scientific approach. We contribute by experimenting a novel method combining elements from several scientific domains like strategic management and system dynamics. We conclude that with language analysis using recurrent neural networks, we were able to predict the success of a news story published on a digital channel in a way that fulfils the 'weak market test' criteria of the constructive approach. A company with whom the model was developed considered it valuable enough to decide to move it from exploration to be further developed and used in real news production.

AI should embody our values: Investigating journalistic values to inform AI technology design

“…These values also serve a democratic function and have been recognised as important in democratic societies [18,35]. As public interest is difficult to delineate [37], journalists were guided by their newsroom's ethos. Likewise while impartiality often involved providing balance, it was also influenced by organisational norms.…”