‍

In this blogpost, we introduce you to the fourth reason why personalizing news is more difficult than in other industries. We will touch base on the techniques of personalization. 

‍

News is deeply tied with democracy and the human aspect really matters.

While one can expect that music recommendation may be fully computed by a machine, it is more sensitive for news. And rightfully so. There is a reason why news is considered as the Fourth Estate, right? 

We believe that the first question of any new personalization use case should be: do we want to apply pure personalization or hybrid personalization? The answer to this question will generally depend on the page where recommendations will be shown. On the homepage, hybrid recommendations will be preferred. On the “sports” category page? Pure personalization will perfectly meet the expectations. 

Now, what is hybrid personalization? There are multiple ways of combining both human and machine intelligence. The easiest way we experienced is to have the human applying its first curation by selecting a short list of articles eligible for recommendation. This can be done ad hoc, or by adding a certain tag to an article. The machine will then apply a second curation by selecting the most relevant articles from the shortlist. 

‍

Human and machine - best friends?

To summarizing, the best method of personalization can be either pure personalization or a hybrid. When we talk about pure personalization, we have two options:

> Collaborative filtering. Recommend an article that a similar reader to already read and liked. 

> Content-based filtering. Recommend an article similar to what you have read and liked. 

‍

A hybrid personalization, then, is a combination of both techniques.

With hybrid personalization you can have the right balance between journalistic curation and technology automation. It helps you give your Editor-in-Chiefs enough deciding power over the diversity of topics and counter-opinions on a subject and to set the constraints on the decisions that artificial intelligence can’t make.

‍

Watch the previous episodes:

• Model Changing Events: The concept of Trendingness
• Short Shelf Life: The Challenge of Real-Time News Recommendations
• How to win agains the filter bubble when personalizing news?