Based on the user's past purchase behavior records, repurchase prediction will predict the possibility of users repurchasing within a specified time in the future by training algorithm models, find a large number of non-brand core groups with high repurchase probability, supplement the business circle with non-brand opportunities, and then focus on these groups with high repurchase probability to improve the brand repurchase rate.
The following figure shows the data and function links.
Data requirements
You need to provide a Behavior Dataset for algorithm model training. Please refer to the Behavior Dataset Sample for the requirements of the behavioral dataset
When making predictions based on an algorithmic model, the range of users and periods that can be predicted depends on the behavioral dataset:
Predictable user range: the users involved in the behavioral dataset used by the algorithm model, and the users in the behavioral dataset must have purchased behavior within the past year, that is, the users who have not purchased behavior within the past year are unpredictable. The last year refers to one year before the latest behavior time in the behavior dataset.
Predictable period range: Based on the last behavior time of the behavior dataset used by the algorithm model, N days in the future from that day. N can be set to 15 to 90. See the example in the following figure.
Process
Purchase the forecast feature pack. For more information, see Sales List and Billing Guide.
Data sources can be connected to AnalyticDB for MySQL. You can analyze only data in spatial or organizational data sources of the ADB 3.0 type. For more information, see Create Spatial Data Source and Organization Data Source Table Authorization.
Prepare a behavior data table as required in Behavior Dataset Sample and Create Behavior Dataset.
Create an algorithm model. After the model is trained, view the training details of the model. For more information about the top 10 training features and model validation, see Model Configuration.
Update the model when the data changes significantly to maintain the accuracy of the population prediction. For more information, see Manually Update.
Create a crowd prediction task based on an existing algorithm model and view and use the prediction results. For more information, see Audience Prediction.