Algorithm creation: completed
The main goal of our 'Triple N' project was the creation of the algorithm that generates actionable notifications to advise newsrooms when to write a follow-up for a certain user need. The research, zero measures, deep dives and growth hacks we did during the project all led up to the moment when our machine learning algorithm would be able to identify user needs in existing stories, and produce tips based on audience behaviour. We are proud to announce that we have currently achieved a 90% level accuracy on the algorithm and that the first batch of notifications is being tested as we speak.
This is what we tasked our data wizards with:
- Create a system of notifications to enable newsrooms to see when a story is ready for a follow up, and identify the optimal way to deliver that story via the user needs approach. We also need insights on business level, to be able to provide practical advice on topics, authors and story performance.
We realised that this was not going to be an easy task, and there were several hurdles to overcome. As far as we could find, categorising content in this way had never been done before. Our CDO discusses this in detail in the interview we published recently.
The three main challenges for the creation of the notifications:
- Creating an algorithm that has to be bombproof in order to be trusted
- Teaching the algorithm to recognise user needs in stories
- Building a notification system when the algorithm isn't 100% accurate yet
In the end, we decided to build notifications for three user need categories rather than the separate six, to prevent it from getting too complicated. These are functional (Update me and Keep me on trend, emotional (Divert me, Inspire me), and contextual (Educate me, Give me perspective). This has proven sufficient, at least at this stage, to identify and trigger accurate notifications.
Some examples of the notifications that are currently being tested:
- After a big spike in traffic this story is still attracting readers. Try creating a follow-up from an alternative user need perspective.
- This story meets your readers' expectations! Try engaging them even more with a ‘Give me perspective’ follow-up.
- This story was very popular, can your team create a follow-up from the ‘Educate me’ angle?
- Twitter is going crazy over this story! How about an explanatory follow-up or an expert’s opinion?
Alternatively, the algorithm can identify the Facebook reactions on a story and provides tips for suitable follow-ups.
The best part is that no two newsrooms will have the same notifications setup. It correlates to their preferred channels, business model, stories, preferences and more. That way it's designed to be the best possible help the newsroom can get to generate more creative articles, more reach and more impact.