First, our proprietary association algorithm ranks by the frequency of paired watches across user sessions, i.e. the top associated slot will be filled by a video that viewers have watched with the current video more than any other content.
Next, our content-based algorithm filters by term frequency and then ranks by the Trending Score, i.e. the top content video will have terms in the title and description that overlap with the current video and has the highest Trending Score.
Finally, backfilling is ranked by the Trending Score, i.e. the top backfilled item will appear after association and content matches and will have the highest Trending Score.