Model Template

The Model Template contains the default settings for training a recommendation model and returning a product recommendation. More specifically, the Model Template values are used when:

  • New model is created.
  • Catalog is loaded.
  • New build is created.
  • Product recommendation is triggered.

Go to Model Template to fill in or edit the Model Template. Note that prior to filling out the template for the first time, make sure that you have completed the steps described in LS Recommend Setup.

In the following, the parameter values and roles of each group in the Model Template are explained in details:

Model data

A recommendation model learns from the given input data. In order to provide good product recommendations, a model needs to be fed with information about the items in the database, called a catalog, and with the history of transactions to predict the interactions between users and items, called usage history.

The parameters listed in the table below are used to query the database for these pieces of information.

Field Description
Distribution Select the Distribution group to define which items are included in a model's catalog. For example, choosing FASHION tells LS Recommend to build a recommendation model only considering items from the FASHION distribution group.
Catalog Time Period Set the length of the default time period for creating a catalog file. A catalog is one of two data input files necessary to create a recommendation model. It is created from the selected Distribution group and the Catalog Time Period. All items in the Distribution group that have transactions during the Catalog Time Period, counting from the day the catalog is created backward in time, are included in a model's catalog.
Include Features In Catalog Indicates if item features are included in a catalog to enhance a recommendation model and to enable cold item recommendation via similarity. Item features are metadata about the products. An example of an item feature is a product group code or an item category code. Features are predefined in the Feature Template (click the Feature Template action). See Features for more information.
Usage History Type Select the type of usage history to train a recommendation model. The values are:
Only Transactions; only transaction data is loaded, which means that user based recommendation cannot be made.
Member And Transaction; all member transactions are considered in addition to non-member sales.
Customer And Transaction; POS customer sales are specially logged in addition to the normal POS transactions.
Usage History Time Period Set the length of the default time period for creating a usage history file. The type of usage data is defined by Usage History Type. The time period starts at the day counting backwards from the day previous to the day that usage data is loaded to a model container.

Recommendation result

A recommendation model returns a list of recommended items. The parameters listed in the table below are used to filter recommendation results.

Field Description
Number of Recommended Items Select the number of items LS Recommend should display in the recommendation list. For example, if the value in this field is 4, LS Recommend displays 4 items in the recommendation list in the POS or in the Commerce product.
Filter With Regard to Stock Indicates if items in the recommendation list are filtered with regard to inventory status.
If you select this check box, only items with stock levels higher than Minimum Item Stock are recommended.
The LS Recommend module uses the web service INVENTORY to calculae stock at the store level. To set up the web service go to Web Service Setup, and fill in Username, Password, and Domain in the Client Credentials FastTab. Note that Override Client Credentials is ignored.
Minimum Item Stock Set the default minimum value for the stock level of items to be filtered. If the stock level of items in the recommendation list are lower than the value in this field, LS Recommend will not display the corresponding items.

Build parameters

LS Recommend supports two additional options to enhance recommendations given sufficient input data. In addition, three advanced parameters are given to tune a model build according to the input data, and thereby improve model predictions.

The build parameters listed in the table below are used as default values when a new build is created manually or by Scheduler jobs. When a new build is created manually, you have an opportunity to change the default parameters, but when builds are created on a schedule the default values are used.

Field Description
Personalized Recommendation Indicates if LS Recommend should make personalized recommendations by default. This type of recommendations is based on specific customer's or member's usage history, which means that items can be recommended to users despite an empty basket. However, if the Usage History Type is Only Transaction, this check box should be clear, because no user specific information is included in the model.
Cold to Cold Item Recommendation Indicates whether the similarity between pairs of cold items, that is new items or items with little sales history, should be computed. If this check box is clear, only similarity between cold and warm items will be computed using the catalog item features. Note: This configuration is only relevant if the Include Features In Catalog check box is selected.
Group Usage History By Indicates how to group the usage history before counting co-occurrences. Basket will group items purchased in the same transaction, and User will group items bought by the same user. If Usage History Type is Only Transaction, there is no difference between these two values.
Minimum Number of Co-occurrence Units A co-occurrence is the number of times two items appear in the same transaction. This value indicates how strict a recommendation model is. If the quality of your co-occurrences is really high, then a higher value (>6) can be suitable. Alternatively, if the recommendation list contains too few recommendations due to a lack of usage history, then a lower value of this parameter is better. Valid values are within the range 3-50, where the default value is 6.
Decay Period in Days Set the length of the decay period to adjust the strength of the usage history in the recommendation model predictions. Events that are older than Decay Period in Days are reduced by half in strength compared to more recent events. The default value is set to 30 days.