Selecting articles to recommend

Joakim Stenberg Updated by Joakim Stenberg

Articles recommended in your widget are automatically selected using a range of algorithms. This is an overview of how we are selecting articles to recommend.

Optimized articles

Goal: Optimize for high CTR while preventing the feeling of click-bait.

You can easily create recommendations optimized for high CTR by only using statistics and this was indeed the starting point for our article recommendations. Quite quickly you'll realize that certain articles (e.g. sex advice and articles containing nude pictures) tend to make this list more often than other types of content. What you end up with is recommendations with a feel of being click-bait and this was also the feedback we got from many customers in the early days.

Today, we are using the similarity of words in articles that we collect to create relations between articles. This creates a more varying set of article recommendations and improves the relevance between what you've just been reading and what we recommend you to read next.

The next issue you might struggle with when using a recommendation model solely based on article CTR statistics is that you need to gain a lot of impressions before you have enough confidence in your CTR score. When Strossle is collecting a new article we want to spend as few impressions as possible to find out if the article will actually attract readers. By comparing articles and the similarity of the words being used, we create predictions of how an article will perform based on historic data. By doing so we prevent a huge waste of impressions while exploring what articles will be best to recommend.

Additionally we always factor in date to keep article recommendations fresh.

Most read articles

Goal: Help readers find popular articles.

Compared to our more advanced default article selection model, this is quite straight forward. Based on reading statistics for articles on your website we recommend the most read articles within a defined time interval. These recommendations tend to be more closely connected to what you are, or have been, promoting on your front page during peak hours and this quality make it a popular recommendation method despite its simplicity and the fact that CTR is often (but not always) lower than our more advanced default article selection method.

Latest articles

Goal: Always promote the most recent articles.

Sometimes you only want to put the latest in front of your readers. For that reason we provide this very simple recommendation technology, only serving what's fresh of the shelf.

Cross website articles

Goal: Increase time spent and reader engagement with websites within our publisher network

To create a rich variety of what sort of articles you promote on your website, it is possible to mix in articles from other websites as well. This is popular especially with publishers hosting a portfolio of different brands but also with publishers who are trying to create synergies through collaboration (e.g. recommending articles within a niche that you might not cover in-depth yourself).

From a reader perspective the results are interesting, typically the length of the total session spans way beyond the time spent on a single website. This way of recommending articles enables the reader to explore new websites directly from the context of our publishers instead of being left to search engines or social networks for this sort of exploration.

Cross website recommendations are only activated when requested by you as a customer. The same underlaying recommendation methods as previously discussed are being used when performing cross website article recommendations.

Mixing articles

It is fully possible to mix what methods to use when selecting articles, in fact we do so quite often to create a variety of optimized, popular and fresh article recommendations.

How did we do?

Elements of an article recommendation

Contact