Base the recommendation on a recommendation key

Recommendations based on algorithms use visitor behavior context to show relevant results in Adobe Target Recommendations activities.

Each algorithm type provides different algorithms appropriate for its type, as shown in the following table:

Each criteria is defined in its own tab. Traffic is split evenly across your different criteria tests. In other words, if you have two criteria, traffic is divided equally between them. If you have two criteria and two designs, traffic is split evenly between the four combinations. You can also specify a percentage of site visitors who see the default content, for comparison. In that case, the specified percentage of visitors sees the default content, and the rest are split between your criteria and design combinations.

For more information about creating criteria and defining its algorithm types and algorithms, see Create criteria.

Different recommendations algorithms lend themselves to placement on different types of pages. Refer to the following sections for more information about each algorithm type and its available algorithms.

Cart-Based cart-based

The Cart-Based algorithm type lets recommend items based on the contents of the visitor’s current cart. The recommendation keys are supplied through mbox parameter cartIds in comma-separated values. Only the first 10 values are considered.

Cart-based recommendation logic is similar to the “Recommended For You” user-based algorithm and to the “People Who Viewed These, Bought Those” and “People Who Bought These, Bought Those” item-based algorithms.

Target uses collaborative filtering techniques to determine similarities for each item in the visitor’s cart, then combines these behavioral similarities across each item to get a merged list.

Target also gives marketers the choice of looking at visitor behavior within a single session or across multiple sessions:

Target makes recommendations for each visitor based on the items in their current cart, regardless whether you look at visitor behavior within a single session or across multiple sessions.

The following algorithms are available with the Cart-Based algorithm type:

People Who Viewed This, Viewed Those

Recommends items that are most often viewed in the same session that the specified item is viewed.

This logic returns other products people viewed after viewing this one; the specified product is not included in the results set.

This logic lets you create additional conversion opportunities by recommending items that other visitors who viewed an item also viewed. For example, visitors who view road bikes on your site might also look at bike helmets, cycling kits, locks, and so forth. You can create a recommendation using this logic that suggests other products help you increase revenue.

If you select this algorithm, you can select the following Recommendations Keys:

People Who Viewed This, Bought Those

Recommends items that are most often purchased in the same session that the specified item is viewed. This criteria returns other products people purchased after viewing this one, the specified product is not included in the results set.

This logic returns other products people purchased after viewing this one; the specified product is not included in the results set.

This logic lets you increase cross-selling opportunities by displaying a recommendation on a product page, for example, that displays items that other visitors who viewed the item purchased. For example if the visitor is viewing a fishing pole, the recommendation could show additional items other visitors purchased, such as tackle boxes, waders, and fishing lures. As visitors browse your site, you provide them with additional purchasing recommendations.

If you select this algorithm, you can select the following Recommendations Keys:

People Who Bought This, Bought That Those

Recommends items that are most often purchased by customers at the same time as the specified item.

This logic returns other products people purchased after buying this one; the specified product is not included in the results set.

This logic lets you increase cross-selling opportunities by displaying a recommendation on a shopping cart summary page, for example, that displays items that other buyers also purchased. For example if the visitor is purchasing a suit, the recommendation could display additional items other visitors purchased along with the suit, such as ties, dress shoes, and cufflinks. As visitors review their purchases, you provide them with additional recommendations.

If you select this algorithm, you can select the following Recommendations Keys:

Popularity-Based

The Popularity-Based algorithm type lets you make recommendations based on the overall popularity of an item across your site or based on the popularity of items within a user’s favorite or most-viewed category, brand, genre, and so forth.

The following algorithms are available with the Popularity-Based algorithm type:

Most Viewed Across the Site most-viewed

The recommendation is determined by the item that has been viewed most often. This is determined by recency/frequency criteria that works as follows:

For example, viewing surfboardA then surfboardB in one session results in A: 10, B: 5. When the session ends, you have A: 5, B: 2.5. If you view the same items in the next session, the values change to A: 15 B: 7.5.

Use this algorithm on general pages, such as home or landing pages and offsite ads.

Most Viewed by Category most-viewed-category

The recommendation is determined by the category that has received the most activity, using the same method used for “most viewed item” except that categories are scored instead of products.

This is determined by recency/frequency criteria that works as follows:

Categories visited for the first time are given 10 points. 5 points are given for subsequent visits to the same category. With each visit, non-current categories that have been viewed before are decremented by 1.

For example, viewing categoryA then categoryB in one session results in A: 9, B: 10. If you view the same items in the next session, the values change to A: 20 B: 9.

Use this algorithm on general pages, such as home or landing pages and offsite ads.

If you select the Most Viewed by Category algorithm, you can select the following Recommendations Keys:

Most Viewed by Item Attribute

Recommends items or media similar to the most-viewed items or media on your site.

This algorithm lets you select which item attribute you want to base the recommendation on, for example, “Name,” or “Brand.”

You then select which profile attributes stored in the visitor’s profile to match, for example “Favorite Brand,” “Last Item Added to Cart,” or “Most Viewed Show.”

Top Sellers Across the Site top-sellers

Displays the items that are included in the most completed orders from across the site. Multiple units of the same item in a single order are counted as one order.

This algorithm lets you create recommendations for top-selling items on your site to increase conversion and revenue. This logic is especially suited for first-time visitors to your site.

Top Sellers by Category

Displays the items that are included in the most completed orders by category. Multiple units of the same item in a single order are counted as one order.

This algorithm lets you create recommendations for top-selling items on your site based on category to increase conversion and revenue. This logic is especially suited for first-time visitors to your site.

If you select the Most Viewed by Category algorithm, you can select the following Recommendations Keys:

Top Sellers by Item Attribute

Recommends items or media similar to the most-purchased items or media on your site.

This algorithm lets you select which item attribute you want to base the recommendation on, for example, “Name,” or “Brand.”

You then select which profile attributes stored in the visitor’s profile to match, for example “Favorite Brand,” “Last Item Added to Cart,” or “Most Viewed Show.”

Top by Analytics Metric

Displays the “Top x” where x is an arbitrary Analytics metric. When using behavioral data from mboxes, you can use Top Sold or Top Viewed (x = “Sold” or x = “Viewed”). If you’re using behavioral data from Adobe Analytics, you could use x = “Cart Adds” or some other Analytics metric.

Item-Based

The Item-Based recommendation type lets you make recommendations based on finding similar items to an item that the user is currently viewing or has recently viewed.

The following algorithms are available with the Item-Based algorithm type:

People Who Viewed This, Viewed That viewed-viewed

Recommends items that are most often viewed in the same session that the specified item is viewed.

This logic returns other products people viewed after viewing this one; the specified product is not included in the results set.

This logic lets you create additional conversion opportunities by recommending items that other visitors who viewed an item also viewed. For example, visitors who view road bikes on your site might also look at bike helmets, cycling kits, locks, and so forth. You can create a recommendation using this logic that suggests other products help you increase revenue.

If you select this algorithm, you can select the following Recommendations Keys:

People Who Viewed This, Bought That viewed-bought

Recommends items that are most often purchased in the same session that the specified item is viewed. This criteria returns other products people purchased after viewing this one, the specified product is not included in the results set.

This logic returns other products people purchased after viewing this one; the specified product is not included in the results set.

This logic lets you increase cross-selling opportunities by displaying a recommendation on a product page, for example, that displays items that other visitors who viewed the item purchased. For example if the visitor is viewing a fishing pole, the recommendation could show additional items other visitors purchased, such as tackle boxes, waders, and fishing lures. As visitors browse your site, you provide them with additional purchasing recommendations.

If you select this algorithm, you can select the following Recommendations Keys:

People Who Bought This, Bought That bought-bought

Recommends items that are most often purchased by customers at the same time as the specified item.

This logic returns other products people purchased after buying this one; the specified product is not included in the results set.

This logic lets you increase cross-selling opportunities by displaying a recommendation on a shopping cart summary page, for example, that displays items that other buyers also purchased. For example if the visitor is purchasing a suit, the recommendation could display additional items other visitors purchased along with the suit, such as ties, dress shoes, and cufflinks. As visitors review their purchases, you provide them with additional recommendations.

If you select this algorithm, you can select the following Recommendations Keys:

Items with Similar Attributes similar-attributes

Recommends items or media similar to items or media based on current page activity or past visitor behavior.

If you select Items/Media with Similar Attributes, you have the option to set content similarity rules.

Using content similarity to generate recommendations is especially effective for new items, which are not likely to show up in recommendations using People Who Viewed This, Viewed That, and other logic based on past behavior. You can also use content similarity to generate useful recommendations for new visitors, who have no past purchases or other historical data.

If you select this algorithm, you can select the following Recommendations Keys:

For more information, see Content Similarity.

User-Based

The User-Based algorithm type lets you make recommendations based on the user’s behavior.

The following algorithms are available with the User-Based algorithm type:

Recently Viewed Items recently-viewed

Uses the visitor’s history (spanning sessions) to present the last x items the visitor has viewed, based on the number of slots in the design.

The Recently Viewed Items algorithm returns result specific to a given environment. If two sites belong to different environments and a visitor switches between the two sites, each site shows only recently viewed items from the appropriate site. If two sites are in the same environment and a visitor switches between the two sites, the visitor sees the same recently viewed items for both sites.

Recently Viewed Items/Media can be filtered so that only items with a particular attribute are displayed.

Possible use-cases include, a multi-national company with multiple businesses might have a visitor view items across multiple digital properties. In this case, you can limit the recently viewed items to display only for the respective property where it was viewed. This prevents recently viewed items from displaying on another digital property’s site.

Use this algorithm on general pages, such as home or landing pages and offsite ads.

Recommended for You recommended-for-you

Recommends items based off each visitor’s browsing, viewing, and purchasing history.

This algorithm lets you deliver personalized content and experiences to both new and returning visitors. The list of recommendations is weighted towards the visitor’s most-recent activity and is updated in-session and become more personalized as the user browses your site.

Both views and purchases are used to determine the recommended items. The specified recommendation key (for example, Current Item) is used to apply any inclusion rule filters you select.

For example, you can:

If you select this algorithm, you can select the following Filtering Keys:

Custom Criteria custom

The Custom Criteria algorithm type lets you make recommendations based on a custom file you upload.

Recommendation is determined by an item that is stored in a visitor’s profile, using either user.x or profile.x attributes.

When this option is selected, the entity.id value must be present in the profile attribute.

When you base recommendations on custom attributes, you must select the custom attribute and then select the recommendation type.

You can perform real-time filtering on top of your own custom criteria output. For example, you can limit your recommended items to only those from a visitor’s favorite category or brand. This gives you the power to combine off-line calculations with real-time filtering.

This functionality means that you can use Target to add personalization on top of your offline calculated recommendations or custom-curated lists. This combines the power of your data scientists and research with Adobe’s tried-and-true delivery, run-time filtering, A/B testing, targeting, reporting, integrations, and more.

With the addition of inclusion rules on Custom criteria, this turns otherwise static recommendations into dynamic recommendations based a visitor’s interests.

Possible use-cases include:

Recommendation keys keys

The following recommendation keys are available from the Recommendation Key drop-down list:

Current Item current-item

The recommendation is determined by the item the visitor is currently viewing.

Recommendations display other items that might interest visitors who are interested in the specified item.

When this option is selected, the entity.id value must be passed as a parameter in the display mbox.

Can be used with the following algorithms:

Use the Current Item recommendations key on your site on:

Last Purchased Item last-purchased

The recommendation is determined by the last item that was purchased by each unique visitor. This is captured automatically, so no values must be passed on the page.

Can be used with the following algorithms:

Use the Last Purchased Item recommendations key on your site on:

Custom recommendations key

You can base recommendations on the value of a custom profile attribute. For example, suppose that you want to display recommended movies based on the movie that a visitor most recently added to his or her queue.

If your custom profile attribute doesn’t directly match to a single entity ID, it is necessary to explain to Recommendations how you want the match to an entity to occur. For example, suppose that you want to display the top selling items from a visitor’s favorite brand.

Last Viewed Item last-viewed

The recommendation is determined by the last item that was viewed by each unique visitor. This is captured automatically, so no values must be passed on the page.

Can be used with the following algorithms:

Use the Last Viewed Item recommendations key on your site on:

Most Viewed Item most-viewed-logic

Displays the items or media that are viewed most often on your site.

This logic lets you display recommendations based on the most-viewed items on your site to increase conversions for other items. For example, a media site could display recommendations on its home page for its most-viewed videos to encourage visitors to watch additional videos.

This recommendation key can be used with the following algorithms:

Current Category current-category

The recommendation is determined by the product category that the visitor is currently viewing.

Recommendations display items in the specified product category.

When this option is selected, the entity.categoryId value must be passed as a parameter to the display mbox.

This recommendation key can be used with the following algorithms:

Use the Current Category recommendations key on your site on:

Favorite Category favorite-category

The recommendation is determined by the visitor’s favorite product category.

Recommendations display items in the specified product category.

When this option is selected, the entity.categoryId value must be passed as a parameter to the display mbox.

This recommendation key can be used with the following algorithms:

Use the Current Category recommendations key on your site on:

Site Affinity site-affinity

Recommends items based on the certainty of a relationship between items. You can configure this criteria to determine how much data is required before a recommendation is presented using the Inclusion Rules slider. For example, if you select very strong, the products with the strongest certainty of a match are recommended.

For example, if you set a very strong affinity and your design includes five items, three of which meet the strength of connection threshold, the two items that do not meet the minimum strength requirements are not displayed in your recommendations and are replaced by your defined backup items. The items with the strongest affinity display first.

For example, an online retailer can recommend items in subsequent visits that a visitor has shown interest in during past sessions. Activity for each visitor’s session is captured to calculate an affinity based on a recency and frequency model. As this visitor returns to your site, site affinity is used to display recommendations based on past actions on your site.

Some customers with diverse product collections and diverse site behaviors might get the best results if they set a weak site affinity.

This logic can be used with the following recommendation keys: