This documentation is for the Classic Editor (Legacy Editor). If you are currently using recommendation schemes created with the Classic editor, you will still be able to use them in your templates until February 13, 2024. However, you will no longer be able to clone them to create new recommendation schemes. We recommend that you try the Photon Editor in Blueshift's expanded Recommendations Studio.
To help you migrate to Blueshift's new Recommendations studio, we have provided a brief mapping of the options from the Classic editor to the Recipes in the new studio.
We support several different types of recommendations, taking into account individual user behavior and aggregate behavior over all users. See below for the common list of recommendations that are available in Recommendation Studio and descriptions on how they are computed. By default, all recommendations are refreshed once daily.
1. Site-wide recommendations
Site-wide recommendations rank catalog items based on all user activity across desktop sites and mobile apps and are based on events sent to Blueshift.
Name | Description | Default parameters |
Top viewed | Ranked list of catalog items sorted by how frequently they were viewed in the last 28 days | lookback_window : 28 days |
Top selling | Ranked list of catalog items sorted by how frequently they were bought in the last 28 days | lookback_window : 28 days |
Top wishlisted/downloaded/.... | Ranked list of catalog items sorted by how frequently they were sent in specific custom event in the last 28 days. For example, event might be add_to_wishlist or download | lookback_window : 28 days |
Note
We can also run the ranking algorithms on an hourly basis for a lookback window that can be anywhere between 2 hours to 24 hours.
For example, in the Top viewed site wide recommendations:
We can run our ranking algorithm each hour, and
You choose the lookback period as the minimum possible setting of 2 hours
In such a case, our platform ranks the top-viewed products in the last two hours. Based on the products' rankings, our platform recommends them to your customers.
Get in touch with us through [email protected] to turn on hourly rankings of catalog items for your account.
2. Category / product attribute recommendations
Category recommendations rank catalog items based on all user activity across desktop sites and mobile apps and are based on events sent to Blueshift for each category. If your category hierarchy is multi level the aggregates are computed at each level of the hierarchy. For example, if your category is similar to the below example, then category recommendations are computed at each of the 3 levels. i.e Travel, Travel > Europe and Travel > Europe > Italy. In Recommendation Studio we define top level to be "Root" category and bottom most level to be "Leaf" category and one immediately above Leaf to be "Parent" category.
Name | Description | Default parameters |
Top viewed | Ranked list of catalog items in this category sorted by how frequently they were viewed in the last 28 days | lookback_window : 28 days |
Top selling | Ranked list of catalog items in this category sorted by how frequently they were bought in the last 28 days | lookback_window : 28 days |
Top wishlisted/downloaded/.... | Ranked list of catalog items in this category sorted by how frequently they were sent in specific custom event in the last 28 days. For example, the event might be add_to_wishlist or download | lookback_window : 28 days |
Lowest price | Ranked list of catalog items in this category sorted by price in ascending order | price > 0 |
Recent additions | Ranked list of catalog items in this category sorted on recency based on start_date attribute i.e (start_date - now) ascending order | start_date IS NOT NULL |
Expiring soon | Ranked list of catalog items in this category sorted on how soon they are expiring i.e (end_date - now) ascending order | end_date IS NOT NULL |
Top discounted | Ranked list of catalog items in this category sorted on discount percentage i.e (msrp-price)/msrp in descending order | price > 0 |
Category recommendations are one type of product attribute recommendations. Thus, the above can be computed for any other product attribute like brand, author, gender, location, language etc.
3. Product recommendations
Product to product recommendations rank items that are related to or likely to be of interest as in next best product to consider. For example, people who view this product also view these other products will rank items based on how frequently they were viewed across all user sessions.
Name | Description | Default parameters |
Next Best Product - view to view | Ranked list of viewed items after viewing this item based on collaborative filtering. The ranking is based on how often this occurs across all user sessions in given window of time. | lookback_window : 91 days |
Next Best Product - view to purchase | Ranked list of purchased items after viewing this item. The ranking is based on how often this occurs across all user sessions in given window of time. | lookback_window : 91 days |
Next Best Product - purchase to purchase | Ranked list of items who go on to purchase other items after purchasing this item. The ranking is based on how often this occurs across all users in given window of time. | lookback_window : 91 days |
Next Best Product - custom event to custom event | Ranked list of items for this custom event after having done this other event. The ranking is based on how often this occurs across all users in given window of time. | lookback_window : 91 days |
Next Best Product - geo distance | Ranked list of catalog items closest in geo distance to this product/item | latitude IS NOT NULL |
Next Best Product - text similarity | Ranked list of catalog items similar to this product/item in text similarity based on title and tags | title IS NOT NULL, tags IS NOT NULL |
4. User feed recommendations
User recommendations are computed based on individual user activity and are ideal for use in daily or weekly newsletters.
Name | Description | Default parameters |
Top timespent items/products | Ranked list of items for each user sorted on how long that user spent time viewing or interacting this item. We use exponential decay to weight the most recent ones above those in the past | lookback_window : 14 days |
Top timespent categories | Ranked list of categories for each user sorted on how long that user spent time viewing or interacting this category. We use exponential decay to weight the most recent ones above those in the past | lookback_window : 14 days |
Top affinity categories | Ranked list of categories for each user sorted on how frequently that user has been interacting with this category across all events | lookback_window : 91 days |
Top products with similar textual content | Top products recommended to a customer based on how similar a product's textual content is to the textual content of a product in the customer's recent catalog activity. Blueshift uses state of the art NLP (Natural Language Processing) algorithms to identify similar products in the catalog for each product based on its textual attributes such as title, category, tags, keywords, and description. | lookback_window : At least 91 days |
Top cross-sell categories | Ranked list of categories for each customer based on the recent purchase activity. Blueshift uses advanced Collaborative Filtering techniques to identify categories of products bought together frequently. | lookback_window : At least 183 days for collaborative filtering and 7 days for user activity |
Top co-browse categories | Ranked list of categories for each customer based on recent browsing history. Blueshift users advanced Collaborative Filtering techniques to identify related categories of products browsed together frequently. | lookback_window :At least 183 days for collaborative filtering and 7 days for user activity |
Blueshift allows a marketer to fully customize each of the above recommendations to suit their campaign needs. Contact your Blueshift Success Manager to set them up in your account. Once set up, they will be available as drop-down options in the Recommendation Studio.
5. Custom imports
You can also import existing recommendations into the platform. This allows you to mix and match imported ones with platform defaults. The feeds are imported via direct file uploads, S3/FTP folders or HTTP end points. We support importing the following types of recommendations:
Type | Description | Comments |
User to Product | Recommendations that go from a single user to an array of products | A CSV file with two columns, user identifier like customer_id or email AND product_ids separated by a delimiter |
Product to Product | Recommendations that go from a single product to an array of products | A CSV file with two columns, source_product_id AND destination_product_ids |