Product recommendation analyzes past purchase records and trains algorithm models to understand the relationships between users and products. This improves operational efficiency, brand conversion rates, and repurchase rates. You can use this feature for the following marketing scenarios:
Use the relationship between users and products for fine-grained operations.
For example, an algorithm can predict a user's preference for each product. During a Double 11 sale, if you want to promote Product A, you can use the algorithm's results to find users who prefer it. You can then create an audience from these users and use the user marketing feature to reach out to them.
Use the relationship between products to recommend related items.
For example, an algorithm predicts a strong link between Product A and Product B. First, you can use the audience filter to find users who have purchased, favorited, or added Product A to their cart. Then, you can use the SMS marketing feature to send text messages about Product B to this audience. The messages can include a short link to the product page or a coupon. If your brand has its own app or online store, you can also use the prediction results to recommend related products on those channels.
The following figure shows the data and feature flow.
Data requirements
To train the algorithm model, you must provide a behavioral dataset and a product tag dataset. For more information about the requirements for these datasets, see Behavioral dataset sample and Product tag dataset sample.
The users, products, and time periods available for prediction depend on the behavioral dataset:
Scope of predictable users: Users from the behavioral dataset who have made a purchase within the last year. The last year is the one-year period before the latest behavior recorded in the dataset. Users with no purchases during this period cannot be included in predictions.
Scope of predictable products: Products from the behavioral dataset that have been purchased at least once in the last year.
Scope of predictable period: The prediction covers a future period of N days, starting from the date of the latest behavior recorded in the dataset. You can set N to a value between 15 and 90. The following figure shows an example.
Usage flow
Purchase the Product Recommendation feature package. For more information, see Pricing List and Billing Guide.
Connect to a data source. This feature can only analyze data from ADB 3.0 workspace data sources or organization data sources. For more information, see Create a workspace data source and Grant permissions on tables in an organization data source.
Prepare a behavioral data table as described in the Behavioral dataset sample. Prepare a product tag data table as described in the Product tag dataset sample. Make sure both tables are in the same ADB 3.0 data source. Then, create a behavioral dataset and create a product tag dataset.
Create an algorithm model. After the model is trained, you can view the training details to understand the relationships between products. For more information, see Model Configuration.
Update the model when your data changes significantly. This helps maintain the accuracy of your product recommendations.
Filter products based on the product tag dataset to create a product pool. This pool defines the candidate products for recommendation. For more information, see Product Pool.
Create a product recommendation task using the existing algorithm model and product pool. You can then view and use the recommendation results. For more information, see Product recommendation task.