Learn about Recommendations

You have hundreds of videos in your library and every week your company produces new content to engage viewers. When viewers come to your site and engage with the video content on the page, how can you address the issue of content discovery: leveraging your video library to keep them engaged to watch more videos after they watched the initial video?

You could create a manual playlist. However, this is a resource-intensive process.

Recommendations solves these problems. Recommendations uses the currently-watched video to automatically suggest videos with the highest likelihood to extend a viewer's watch session. Our solution uses a proprietary algorithm that takes into account sets of videos watched consecutively, sets of videos completed consecutively, sets of videos watched for various time intervals, the trending score of the videos, and the metadata associated with the video library (titles, descriptions).

When using Recommendations, a content editor only needs to select the initial video that a viewer sees. Then, JWP's algorithms create a playlist with no additional editorial work.

Benefits

By turning on Recommendations, publishers achieved the following:

Increased Engagement: Publishers generate 88% follow-on plays per initial play, an increase of 30% over trending-now playlists.
Increased Ad Inventory: Publishers generate 79% follow-on ad impressions per initial play.
Time Savings: Editors do not need to create manual playlists. They only need to choose a seed video and JWP will generate the playlist with videos with the highest likelihood for maximizing time watched.
Resource Savings: Implementation consists of a few clicks in your JWP dashboard, which is fully integrated into the publisher workflow.

Recommendations explained

Recommendations uses three layers of recommendation methodologies to increase the engagement of your viewers: association, content similarity, and trending.

Association layer

In the association layer, JWP uses a proprietary Pairwise Empirical Engagement Rate (PEER) model to find videos that are regularly watched and completed together by your viewers. This model selects videos that have the highest likelihood of being watched in a consecutive manner based on the following data:

Sets of videos watched together across viewer segments
Sets of videos completed together across viewer segments
Set of videos watched for various time intervals
Order in which videos are watched together

The PEER model also addresses the symmetry problem within recommendations. A simple way to think about this problem is to imagine an e-commerce use case. For example, most shoppers that buy beds first also buy bedsheets. However, most shoppers that first buy bedsheets do not buy beds. Just because the first set of customers bought beds and bedsheets together does not necessarily mean that beds should be recommended to buyers of bedsheets.

The same e-commerce problem exists in video recommendations. Just because a set of viewers watched Videos A and B does not necessarily mean that we recommend Video A every time a viewer watched Video B. The PEER model addresses this issue of symmetry and also takes into account the order in which videos are watched together.

Content similarity layer

When our engine cannot build sufficient associations between videos, the Recommendations engine uses metadata -- such as video title, description -- to compute the semantic similarity between two videos.

Trending layer

In the event that neither the association layer nor the content layer generates strong signals, the Recommendations engine falls back to trending videos based on an Exponentially Weighted Moving Average (EWMA) of plays to highlight recently popular videos.