This topic provides a high-level description of the capabilities of Recommendations AI. For a description of the process of implementing Recommendations AI, see Implementing a Recommendations AI solution.
Recommendations AI enables you to build high quality personalized product recommendation systems without requiring a high level of expertise in machine learning, systems design, or operations. Leveraging your website's catalog products and user behavior, Recommendations AI builds a recommendation model specific to your company, including optional add-on features such as result diversity, selecting whether you are optimizing for CTR/CVR/Revenue, and shopping feed integration (all available by request to the support team. You can then request recommendations for other catalog products to display to your users.
In order to build recommendation machine learning models, Recommendations AI needs two sets of information:
Product catalog: Information of the products sold to customers. This includes the product title, description, in stock availability, pricing, and so on.
User events: End user behavior on your website. This includes users searching for, viewing, or purchasing a specific item, your website showing users a list of products, and so on.
The Recommendations AI
The Recommendations AI provides capabilities for two tasks:
Data Ingestion: You can upload and manage product catalog information and user event logs for your websites. Recommendations AI uses this information to train and update recommendation models.
Prediction: You can request recommendations based on your product catalog and user event logs.
Implementing a Recommendations AI solution
To integrate Recommendations AI into your website, follow these steps:
|1. Set up a Google Cloud Platform project||
To use Recommendations AI, you must create a Google Cloud Platform (GCP) project and create authentication credentials including an API key and an OAuth token (either using a user account or a service account) to access the project.
You can only host one product catalog per GCP project.
If you want to create separate environments for testing and production, set up two GCP projects, one for testing and another for production. Because the two projects are separate, you will need to perform all of the setup steps for both projects, including importing your product catalog information.
|2. Import your product catalog||
You can add items to your Recommendations AI product catalog
individually by using the
You will get the best results for recommendations if you provide highly-detailed catalog information and keep your imported product catalog as up-to-date as possible. Detailed and accurate catalog information results in a better recommendation model.
|3. Record user events||
After you have finished importing your product catalog, you are ready to start recording user events. User events track user actions such as clicking on a product, adding an item to a shopping cart, purchasing an item, and so on. Recommendations AI relies on user event data in order to generate personalized recommendations. User events need to be ingested in real time to accurately reflect the behavior of your users.
You have several options to record user events:
|4. Request recommendations||
After you have created your product catalog and are logging user events, you can now request recommendations from Recommendations AI for specific users using your website.
You can request a recommendation by calling the
|5. Evaluate your model||
You can associate recommendations and user events and Recommendations AI provides reporting of metrics to help you determine how incorporating the recommendations is affecting your business.
You can view recommendation metrics for your project in the Dashboard tab of the Recommendations AI Console.
|6. Set up an A/B experiment (Optional)||
You can compare the performance of your website with Recommendations AI recommendations to a baseline version of your website without Recommendations AI recommendations. To compare versions of your website, set up an A/B experiment that randomly partitions a subset of your users into control and experimental groups. The control group sees your baseline website; the experimental group sees your website with Recommendations AI recommendations.