Preview recommendations

Data processed by Discovery for Media will be governed by the standard Google Cloud Platform Terms of Service and the Pre-GA Offerings Terms of the Google Cloud Platform Service Specific Terms. Pre-GA products might have limited support, and changes to pre-GA products and features might not be compatible with other pre-GA versions. For more information, see the launch stage descriptions.

Discovery for Media is a "Discovery Solutions" Service and is subject to the Discovery Solutions Consent Addendum to the Google Cloud Platform Agreement.

This page describes how to you can preview recommendations.

You should preview some recommendations before putting the serving config into production.

To preview recommendations, you'll specify a serving config, which has a trained model associated with it. The preview recommendations feature can return a list of documents that would be recommended to a user.

Previewing allows you to evaluate the following:

• The quality of the data in the datastore. If your data is incomplete, for example missing date or category information, the quality of the recommendations will be poor.

• The quality of the user events

• The completeness of the model's training and tuning

• The choice of optimization objective for the model

• The effectiveness of any serving config options that you applied for diversification or demotion

Before You Begin

Before you can preview recommendations, you need a trained recommendation (model) and one or more serving configs.

Preview a recommendation

When you preview a recommendation, you see the list of documents that would be presented to a user, based on the user events associated with the user. Additionally, in the case of Others You May Like and More Like This models, you see the document on the detail page where the recommended documents are presented.

To preview a recommendation:

1. Go to the Discovery Engine Evaluate page in the Google Cloud console.
   
   Go to the Evaluate page

2. Select the serving config that you want to evaluate.

3. Optional: Enter a visitor ID to preview recommendations for that user.

   If you don't enter a visitor ID, you receive recommendations that would be presented to a user who has no history of interacting with your site.

4. If the Associated documents section is shown, click Add an item and enter a document ID to get recommendations for that document.

   This document represents the document on a detail page where the user views the Others You May Like recommendation.

   For more information, see the following section About associated documents.

5. Click Prediction preview to see the prediction results.

About associated documents

The Associated documents section is available if your serving config uses the Others You May Like model. This model is applied to the Detail page. If you add multiple documents, you are representing a (less common) situation where your Detail page shows more than one document.

If you provide a document ID that isn't in the datastore, you get an error message and the document is ignored in the prediction results. For example, if you specify an invalid document ID and no visitor ID, the recommendation result is a list of the generally popular documents.

Next steps

If you are satisfied with the quality of recommendations that you previewed:

• Get recommendations.

If you are unhappy with the quality of recommendations that you previewed, try one or both of the following:

• Edit the serving config.
• Tune the model.