With news consumption becoming increasingly digital, news platforms have to work hard to transfer and retain their user base. A proven method of maintaining and increasing the user base of any digital platform is to apply personalization techniques. Currently, news platforms tend to prefer manually selecting articles that should be promoted on the front page. This is partly because they have always worked like this, but also because they are simply unaware of the real world effects of applying personalization techniques to their news feeds. As these news feeds are becoming their primary interaction with their customers, their hesitation to try this technology is understandable. With this research we aim to take some of this hesitation away, by providing some valuable insights into the effects of content-based filtering on news feeds.
This blog provides a look into research conducted for my bachelor thesis. It is written in collaboration with Max Knobbout, Lead Artificial Intelligence at Triple.
We will explain the technical concept, the experiment setup and the most important conclusions. The original research was done using a dataset containing articles from a Dutch regional news platform. In the technical analysis we will be using an open source alternative.
Table of Contents
Technical Concept: Recommendation as Classification
In this section we will explain how we can view the problem of recommending relevant news articles as a classification problem. Generally speaking, there are two families of recommendation algorithms:
#news #classification #recommendations #data-science #data analysis
