Goal

Increase the amount of articles a user reads by providing relevant “read more” suggestions”.

Case

Publishers are able to generate large amounts of content everyday. Too much content for any person to read. This leads to a major problem: how do we match a person with relevant content?

A newspaper had a “read more” section on their article detail pages to entice readers to read more. The problem is that there is only room for a few “read more” articles even though hundreds of articles are written.

This creates a problem, which articles should we put in the “read more” section?

By default a simple ranking system was used by the newspaper’s CMS-platform. This basically was a simple “what article is read most” algorithm. Simple conditions like “did the user already read the article” or “is the article the article that is being read now” were not taken into consideration.

This could be improved so we started by testing 2 algorithms against the default algorithm.

The first variant is an algorithm based on the Topic-Sensitive PageRank algorithm. This algorithm uses past interactions between users and articles to predict which articles are often read together. Based on the articles that a user already read, it gives a recommendation on what articles the user could read next.

Variant 2 is the same algorithm as variant 1, with the addition that this variant gives more weight to the article the user is currently reading.

Outcome

Variant 2 was the winning variant. There was an increase of 4,4% more pageviews across all users and an increase of 69,8% more clicks on the “read more” section.