Configure Product Recommendation Model
The Configuration step allows you to select from several recommendation types and define their thresholds. Individual types of product recommendations are organized into groups and you can easily turn them on/off using the checkboxes. Once you enable a recommendation type, its parameters will appear.
Recommendation Types
To use an individual type, use its main checkbox to enable it. Each type can be selected independently.
Bought Together
After you enable recommendations for products usually bought together, you need to fill in the following options:
Minimum Number of Product-CoProduct Transactions – Minimum number of transactions observations to consider for Product Specific recommendations.
Example: Product A → Product B transaction count 1, minimum threshold set to 3, so at least 3 transactions with Product A and Product B are required, otherwise this recommendation Product A → Product B is excluded.Maximum Number of Recommendations Per Product – Maximum number of recommendations to generate for each product (Top N recommendations Stored).
Customer Based Recommendations
In customer based recommendations, you can use two recommendation types: Frequently Purchased and Others Buy. You need to fill in the following options:
First, identify the customer by filling in Customer Name and Customer ID.
Frequently Purchased
When defining Frequently Purchased:
Minimum Number of Customer-Product Transactions – Minimum number of transactions observations to consider for Customer Specific recommendation.
Example: Customer C1 → Product A transaction count 1, minimum threshold set to 3, so at least 3 transactions with Customer C1 and Product B are required, otherwise this recommendation Customer C1 → Product A is excluded.Maximum Number of Recommendations Per Customer – Maximum number of recommendations to generate for each customer (Top N recommendations Stored).
Others Buy
When defining Others Buy:
Customer Segment Labels Source – Defines if the segments are coming from an existing field in transactions (= transaction source) or are generated by the model.
Transaction Source Customer Segment Label – Defines dimension in Transaction Source to use as a segment label.
Minimum Number of Segment-Product Transactions – Minimum number of transactions observations to consider for Customer Segment Specific recommendations.
Example: Customer Segment S1 → Product A transaction count 1, minimum threshold set to 3, so at least 3 transactions with Customer Segment S1 and Product B are required, otherwise Customer Segment S1 → Product A recommendation is excluded.Maximum Number of Recommendations Per Segment – Maximum number of recommendations to generate for each customer segment (Top N recommendations Stored).
Product Similarity Based Recommendations
In product similarity based recommendations, you can use these recommendation types: Up-sell, Down-sell, Similar Products. You need to fill in the following options:
First, select the Product Similarity model.
Up-sell / Down-sell
When defining Up-sell / Down-sell:
Product Similarity Threshold – Defines what is the minimum product similarity that should be considered.
Priority – Select what should be given higher priority:
Higher Margin (for Up-sell) / Lower Price (for Down-sell)
Product Similarity
Combined Margin and Similarity (Score)
Maximum Number of Recommendations – Maximum number of product recommendations to generate.
Similar Products
When defining Similar Products:
Product Similarity Threshold – Defines what is the minimum product similarity that should be considered.
Maximum Number of Recommendations – Maximum number of product recommendations to generate.
Similar from Another Brand
When defining Similar from Another Brand:
Brand / Brand to Recommend – Allows you to set up mapping between brands. If there is a specific brand (selected in Brand), then a similar product from Brand to Recommend will be recommended.
Product Similarity Threshold – Defines what is the minimum product similarity that should be considered.
Maximum Number of Recommendations – Maximum number of product recommendations to generate.
Overall Weights for Product to Product Recommendations
The last section on this screen “Overall Weights” allows you to define the weight of a particular recommendation type. By default it is 1.
Once weight is defined, it is combined into a score for each product recommendation:
OverallScore = sum(weight * each score) / sum(weight)
You can check out this score in Model Tables:
When finished with all configurations, click Save Model, and continue to the next step Product Recommendations.