Configure AI engine for Recommendations

Before you can use AI recommendations, you must prepare an item feed which will be the source of items.

The process consists of the following steps:

1. Adding product feed
2. Selecting a product feed
3. Selecting attributes for preview
4. Grouping feed attributes
5. Selecting a recommendation type
6. Selecting response attributes
7. Selecting filterable attributes
8. Defining the item link
9. Selecting training attributes
10. Selecting attributes increasing the item variety

After the configuration and the model training, you can monitor the status. When the model is active, you can create a recommendation.

Adding item feed

The first step is selecting the feed from which items will be sourced. You can either select a catalog that contains a feed or use Google Merchant Feed. Make sure, the item feed includes the following attributes:

• title - The name of the item
• category - The category of the item

You can read about requirements for item feed:

• uploaded to a catalog
• pulled from an URL
  Grouping feed attributes is available for this feed type.

If you previously configured a feed which you want to use now, you can omit this step.

1. Go to Settings > AI engine configuration.
2. Click Add feed.
   Result: A pop-up appears.

   

3. Select the product feed you want to use.

   • Catalog

     1. On the pop-up, select the type of catalog:
        • Data catalog
        • Metadata catalog
     2. From the dropdown list, select a catalog.

        

     3. Confirm by clicking Apply.
        Result: An item feed based on the contents of the selected catalog appears on the list in Settings > AI Engine configuration.

   • Google Merchant

     Tip: Include the Last-Modified header to enable detection of whether the product feed import is necessary, this will help you reduce data transmission costs.
     1. Provide the following information:
        • the link to the Google Merchant feed (Feed link),
        • the name of the field (Feed name),
          this name will be visible on the list of item feeds in Synerise in Settings > AI Engine Configuration
        • the frequency of pulling updates from the feed to Synerise (Interval),
        • authentication type (Authentication type),
          • username and password (Basic authentication type)
          • username, password, token type, URL (Bearer authentication type)

        

     2. Proceed to the next step by clicking Next.
     3. If you want to group feed attributes, enable the toggle. Further grouping settings are defined after adding the feed.
        Grouping feed attributes refers to organizing the attributes or fields within an item feed into logical groups. Instead of treating each product variant — such as different sizes or colors — as a separate item in the feed, all variants of a product are combined and grouped into a single item.

        

        WARNING: This action is irreversible after applying.
     4. Confirm by clicking Apply.
        Result: The feed appears on the list in Settings > AI Engine configuration.

Selecting item feed

1. In Settings > AI Engine Configuration, on the list of feeds, click the feed you added according to the Select product feed procedure.
   The configuration form opens. In the Item catalog section, the feed you are configuring is selected automatically and you can proceed to the next part of the configuration.

   

Selecting attributes for preview

You can define attributes whose values will appear in the Synerise platform preview when you test the recommendation or the search settings.

1. On the Attributes for preview tab, click Show.
2. In the Response attribute column, from the lists select the attributes that contain the data in the Item attributes column.

Example: If the data source stores the item title in the itemTitle attribute, choose itemTitle: on the right, select the itemTitle attribute as the pair for Title from the left column.

Result: Values for the attributes that are chosen in the Response attribute column will be shown for products when previewing them in the Synerise platform.

Grouping feed attributes

Grouping feed attributes organizes attributes or fields within an item feed into logical groups. Instead of treating each product variant — such as different sizes or colors — as a separate item in the feed, all variants of a product are combined and grouped into a single item.

During this grouping process, a “base” product is selected as representative. This is done by first gathering all products that share the same itemGroupId, then grouping them by category, and finally selecting the first product from the largest category group to serve as the base.

When adding such a feed to Synerise, you must enable the grouping of feed attributes. Please note that enabling grouping is irreversible.

1. In the Grouping feed attributes section, click Show.
   Result: Grouping rules are shown. The system automatically prepares grouping rules.

2. Verify the proposed grouping rules.
   When configuring the grouping of feed attributes, you must map source fields to target fields using one of the following aggregators. Each aggregator determines how values from multiple items are combined:

   Aggregator	Description	Example	Aggregated Result
   or	Performs a logical OR on boolean values. Returns true if any value is true, otherwise false.	true, false, false	true
   arrayDistinct	Aggregates unique, non-null values into an array, removing duplicates.	"41", "42", "41", "42"	["41", "42"]
   arrayFlatten	Flattens nested arrays into a single array combining all elements.	["123", "456"], ["456", "789"]	["123", "456", "456", "789"]
   array	Aggregates values into an array, preserving order.	"41", "42"	["41", "42"]
   single	Selects the first non-null value from the list. For the availability attribute, the aggregated result will be a boolean value.	"PROD123", "PROD124"	["PROD123"]

3. To add your own rule, click Add rule.

   1. In the Source filed dropdown, select an item attribute from the feed pulled from Google Merchant.
   2. From the Aggregator dropdown list, select the aggregator based on which attribute values will be combined.
   3. In the Target field dropdown list, select the attribute which will hold the aggregated values. You can create an attribute.

4. Confirm by clicking Apply.

Selecting recommendation types and default filters

Select the recommendation models you want to enable for the selected item feed. Optionally, you can define default filters (global item and distinct filters) for the recommendations created based on the recommendation model for a specific item feed. These filters include:

• Item filter
  • This is a global item filter whose conditions an item must meet to be included in the recommendation of a specific type.
  • You can enable this filter in the settings of a recommendation campaign through the Apply Items Global Filters toggle while adding slots. The filter will then work together with the filters in the recommendation slot.
  • If you change the filter settings, the changes will be applied automatically to all recommendations based on the specific recommendation model for which the Apply Items Global Filters option is enabled.
• Distinct filters
  • This is a global distinct filter which helps increase the variety of items in the recommendation of a specific type. Applying this filter limits the number of items with the same value of an attribute (such as, brand, color, shape, category).
  • The choice of attributes for selection in this filter is defined in the Attributes for distinct filters section in AI Engine Configuration.
  • You can enable this filter in the settings of a recommendation campaign through the Apply Global Distinct Filters toggle while adding slots. When you apply it there, it will override the distinct filter for a slot.
  • If you change the filter settings, the changes will be applied automatically to all recommendations based on the specific recommendation model for which the Apply Global Distinct Filters option is enabled.
  • For all recommendation types except for Last seen, the engine considers up to 1000 items with the highest score that match the recommendation type. For example, if you selected the Cross-sell type, the engine analyzes up to 1000 items that match the cross-sell recommendation type, and then selects the number of items you chose to include in the slot.
  • For the Last seen recommendation type, the engine considers the last 100 page visit events. Based on the data from these events, the engine selects the number of items you chose to include in the slot.

In the Recommendation models section:

1. On the recommendation model you want to enable, click Show.

2. Switch the Model enabled toggle on.

3. Optionally, define default filters for the recommendation model. These are filters which are automatically applied to the recommendation, in the Default Filters section:

   1. In Item filter, click Add.

      1. From the dropdown list, select an item attribute.
      2. Use a logical operator.
      3. Enter or select a value of the item attribute.
      4. To add more conditions, click Add another and repeat steps I-IV.
      5. Confirm by clicking Apply.

   2. In Distinct filters, click Add.
      

      Note: You can read how to build conditions in the distinct filters for the category attribute here.

      

      1. In the pop-up header, enter the name of the filter.
      2. In the Show me only field, enter the number of items whose attribute values can be the same.
      3. From the Choose attribute dropdown list, select the attribute.
         You can use each attribute only once. This means that if you add multiple filters, any attribute that has already been used will no longer be available for selection.
      4. Confirm by clicking Apply.
      5. To add the next distinct filter, click Add and repeat the steps.
      6. If you want the distinct filter to supplement the slot with non-matching items in case not enough matching items are found, enable the Mark filter as elastic option.

4. Repeat steps 1 to 4 for other recommendation models you want to select.

5. Confirm by clicking Apply.
   Result: The AI model is trained. The Default Filters are disabled in the settings of a recommendation campaign by default.

Selecting response attributes

A response attribute is an attribute of the item that is returned in the response to a recommendation request.

1. In the Response attributes tab, click Show.

2. Click Select attributes.

3. Select the checkboxes next to the attributes which you want to include in the response to a recommendation request.
   There are two types of attributes:

   • Textual attributes: String-type attributes, for example, a color, an item name, a brand name, fabric, pattern, and so on.
   • Range attributes: Attributes that can have numerical values within a selected range, such as size, price, width, length, and so on.
   Tip: To limit the volume of data sent to your applications and sites, you should only select the attributes which you plan to use.

4. Confirm by clicking Apply.

Selecting filterable attributes

You can define the attributes which you can use later to filter recommendations results as well as allow you to filter items in the settings of the Propensity predictions. The attributes also become available in the Decision Hub.

1. On the Filterable attributes tab, click Show.
2. Click Select attributes.
3. Tick the checkboxes next to the attributes which you want to use for filtering the recommendation results.
   Recommended: Don’t use title or description as filterable attributes. Using these attributes as filterable has a negative impact on performance.
4. Confirm by clicking Apply.

Defining the item link

An item link is an attribute of an item to which Synerise’s UTM parameters are added.

1. On the Definition of item link tab, click Show.
2. From the Attribute dropdown list, select an attribute that is the item link. You can select more than one attribute.
   
   Tip: When you use one workspace for multiple language versions, you can select several attributes as an item link. This lets you analyze the statistics such as item clicks or visits across various website language versions.
3. Confirm by clicking Apply.
   Result: Thesnrai, snr_content, and snr_id parameters are added to the URL of the item. For example: https://www.exemplary-shop.com./winter-shoes-camelbrown.html?snrai_campaign=QWERTY[…]e=&snrai_content=&snrai_id=123456789010305

Selecting training attributes

Select the attributes which will be used for training for the Similar items and Visual similarity models, so they can produce relevant results.

• For the Similar items and Item comparison models, the title attribute is set at default.
• For the Visual similarity model, the default attributes are: image link, additional image links, and availability. These attributes are necessary as the location of the data and images are required.

We recommend selecting short attributes such as title, color, brand.

1. On the Training attributes tab, click Show.
2. Click Select attributes.
3. Tick the checkboxes next to the attributes which you want to display to the customers.
   Note:

   There are two types of attributes:

   • Textual attributes: String-type attributes, for example, a color, an item name, a brand name, fabric, pattern, and so on.
   • Range attributes: These are all attributes that can have numerical values within a selected range, such as size, price, width, length, and so on.
4. Confirm by clicking Apply.

Selecting attributes to increase item variety

By using distinct filters, you can define the number of items with the same value of an attribute (for example, a brand) that can be displayed in the recommendation frame.

1. On the Attributes for distinct filters tab, click Show.
2. Click Select attributes.
3. On the pop up, from the list, choose up to 5 attributes.
4. Click Apply.
5. Confirm the settings of the tab by clicking Apply.
   Result: These attributes are available in the Distinct filter while creating a campaign and in the Item attributes dropdown in the the slot configuration for the Section recommendations and the Attribute recommendations.

After the configuration, you can monitor the status. When the model is active, you can create a recommendation.

Removing item feed

You can remove item feeds from the list. While removing item feeds, any Synerise objects that rely on the feed as a source of information (such as predictions, recommendations, AI search) will be automatically disabled. Prior to confirming the removal operation, you will be presented with a list of objects that are currently using the feed to be removed. The feed will be removed after 7 days from confirming the action. During this transitional period, the feed and the objects associated with it will remain active. In the case of a feed within a catalog, the contents of the catalog will be deleted, while the catalog itself will be retained.

1. Go to Settings > AI engine configuration.
2. Next to the name of the feed you want to remove, click
3. From the dropdown list, select Delete.
   Result: A pop-up appears with the list of objects that are currently using the feed.
4. To confirm the operation, click Yes, delete feed.