Basic recommendation algorithms are statistic based and need only the ID of a product and the ID of a user. This information is provided by the event tracking as described in the chapter 8. Event Types. There are several use cases where the recommendation engine requires additional information linked to the products. Most important cases are:
- Filtering results based on the category the product is located in (see 7. Category Filter for more information)
- Filtering results based on price and product availability (see 9. General Filters)
- Sub-models based on custom attributes (see 10. Sub-Models).
- An additional export/import interface is designed to provide editor based models (see 5. Recommendation Models).
There are three different ways to import data into the recommendation system.
- In-Event. In the event tracking process, where related data is attached to click event (e.g. category path information)
- Push interface. Classic HTTP POST based import. It is suitable to upload single product or editor lists.
- Pull interface. The recommendation engine loads an exported file from the specified location in the background. It was designed to upload big portions of information, for example a weekly update of the whole product catalog.
Following is the overview table for all import/export interfaces.
|Available for Basic edition|
|Updates the category information of products|
|Updates prices and custom attributes|
|Batch processing for big amount of data (more than 100 products)|
|Export from the recommendation engine|
|Editor based models update|
|Test tool with format validation and error handling|
|Deletes the given item from the item store|
Specification for Import/Export interfaces one can find in the developer guide