Personalization Algorithm Use Cases
Serve users optimal content with personalization algorithms. From your trending content to items centered around your users’ specific interests, algorithms help you curate the most relevant content to deliver in your messaging and on your site.
The following table shows the algorithms grouped by type, including an explanation of how the algorithm works and when to use it.Note: If there is not enough content to meet the personalization and zephyr settings, the system will fall back to recommending content using the popular algorithm, ignoring any filtering rules and cancel/assert statements.
|Name||How they work||When to use these algorithms|
|Popular Trending||Our ranking algorithms are based on overall popularity, showing content items which have the most pageviews or purchases.
Popular shows the most popular items of all time, based on purchases (or pageviews, if no purchase data is available), while trending shows items that have had the greatest increase in popularity in the past week (based on purchases/pageviews). While popular is based on the absolute number of purchases/pageviews, trending looks at the relative change in purchases/pageviews over the 7 day timeframe.
|Popular or trending can provide recommendations for an audience where you have limited individual user engagement data (email or onsite). Use trending rather than popular if you want to improve customer discovery across a wider range of items across in your content library.
These algorithms also work well as a fallback, when using algorithms that rely on individual user engagement data (context, purchased or viewed). Popular/trending work best when:
|Context Purchased Viewed||These algorithms are based on wisdom of the crowd to find similar or related content; providing highly personalized recommendations based on pageview and purchase data from similar users. For a given item, context is based on views/purchases from all time, returning the items from your content library which are most often viewed or purchased by users who viewed or purchased the original item. For a given user, viewed or purchased take into account the given user’s recent view/purchase history, returning a set of recommendations based on other items that similar users also viewed/purchased.||Use context for transaction-specific onsite or email recommendations, to unknown users / an audience where you have limited individual user engagement data.
Use purchased for onsite or email recommendations to an audience of customers with an established purchase history.
Use viewed for onsite or email recommendations to help your audience discover relevant products they wouldn’t have found on their own. Your audience must have at least two pageviews.
Context/purchased/viewed work best when:
|Interest||The interest algorithm filters content based on your content tags, returning items in the selected feed with tags that match the user’s interest profile. This algorithm uses weighted scoring to calculate the user’s interest in an item compared to other users’ interest in the item.||Interest was the original algorithm. Use interest if you only want to show recommendations from a specific subset of items from your content library.
To show personalized recommendations (i.e. the most relevant content from across your entire content library); use context, purchased or viewed (rather than interest) for best results.Interest works best when:
|Random||A randomized selection of content from a data feed.||Use random as a control when A/B testing to understand the effectiveness of personalization algorithms.|
|Custom||Use your own custom algorithm to serve recommendations to your users. Call in specific Content Library items — by URL, SKU, title, or content_id — which your own algorithm may have preselected for the given user and added to their user profile as an array of content identifiers.*||Use custom to support your own / third party recommendation algorithms. If you simply want to recommend a list of specific items from your content library, use interest by filtering a Data Feed in Recommendations (rather than using custom) for best results.|