Our system looks at both metadata and the content of a post to decide what is related.
For instance, we look at metadata like tags and categories.
We also pull out semantically important terms from the body of posts. Our method of doing this is not limited to English and we support any language.
We then use those features and important terms to compute the strength of relationships between various pieces of content in order to make relevant recommendations.
For some posts, we'll have multiple possible sets of recommendations, and we automatically A/B test these in front of readers to figure out the best strategy for that particular post.
The system constantly recomputes recommendations as new content and reader feedback enters the system.
For enterprise and professional clients, we can also integrate videos and products into the related section, showing them only when they are actually related.