Contextly learns about a site's content using both meta-data and the content of a site's stories, videos and products.
We then further learn about a site by capturing and analyzing how readers interact with the site.
All of this work happens on our servers that are equipped to calculation millions or even tens of millions of relationships between the content on your site. This allows publishers to have great recommendations without slowing down their site or taxing their own databases, which are not built to do these kinds of calcualations.
Once we learn a site, we begin to create a two main types of recommendations:
1) Related: We use a proprietary recommendation system that uses meta-data (such as tags and categories), along with text analysis to find the best related content. For sites that have text and video content, we can include video recommendations for text stories.
2) Explore: Readers aren't always in deep-dive mode, so we use the explore section to show off the best from a publisher's site.
This section contains a blend of 3 strategies:
a) Most popular: We calculate the most popular content from the site, looking over the last 30 days, but with heavier weighting towards stories in the last 7 days.
b) Evergreen: We are able to find which of your older stories are still valuable to readers long after they've fallen off the homepage. We keep these alive and they often perform even better than Most Popular.
We also include curation tools for our WordPress and Drupal clients so that writers and editors can choose related posts, if they want to. They can also use our tools to create in-story sidebars and to easily create links in the body of stories to their previous content, which is great for SEO.