AWS Personalize Filter Recommendations by ItemID

Question

In AWS Personalize, I have the "RELATED_ITEMS" recipe
We trained the model on the whole items we have for all menus

But the menu's items will be different every week, therfore we cannot recommend items from outside the current week items

So how to tell the model to recommend from specific items I pass to him, for example filter by ItemID, without retrain the model every week?

Anna_G · Answer

Hi,
For Amazon Personalize, the best way to achieve this for your use-case would be to have a field in your idem schema, for example {available = "yes"} then create a filter ( [Filtering](https://docs.aws.amazon.com/personalize/latest/dg/filter-expressions.html) ) when you use the getRecommendations API to only return items with  {available = "yes"}. You can change with items are available each week using the [putItems API](https://docs.aws.amazon.com/personalize/latest/dg/API_UBS_PutItems.html). This way you can control what items are returned inthe recommendations without retraining the model each time.

I still recommend retraining the model frequently to make sure the latest interactions are taken into account and your recommendations are as good as possible.

This blog shows a sample of how to use filters and promotions: https://aws.amazon.com/blogs/machine-learning/customize-your-recommendations-by-promoting-specific-items-using-business-rules-with-amazon-personalize/

AWS Personalize Filter Recommendations by ItemID

Relevant content