Artificial Intelligence Recommendation:Pricing

This topic describes the billable items of PAI-Rec, as well as the billable items of related cloud resources.

Billable item 1: the recommendation engine, experiment platform and reports, and services such as recommendation solution customization

The prices of resources may vary with regions. Therefore, the prices of the recommendation engine also vary with regions.

Standard Edition provides the following capabilities:

• Recommendation engine

  The recommendation engine combines different modules such as recall, filtering, coarse ranking, fine ranking, and cold start, to generate recommendation results based on user requests.

• Feature consistency validation

  This capability records the online user features, item features, and context features in the recommendation service based on the online requests of the recommendation system, transforms these features, and then imports generated features into models to generate scores. This capability records the corresponding offline features, transforms these features, and imports generated features into models to generate scores. Then, this capability compares the scores generated for the online features and offline features. If the scores are inconsistent, this capability identifies the features with different feature values. This way, the inconsistency between online features and offline features is quickly identified. This ensures the consistency of online and offline feature storage, reading, and transformation.

• Experimental platform

  You can use this platform to manage the recall, filtering, ranking, and re-ranking parameters of recommendation scenarios, and perform tokenization experiments by adjusting experiment configuration parameters. The experiment platform provides capabilities such as experiment traffic management, experiment metric registration, and comparison of experiment reports on a daily or hourly basis.

Advanced Edition adds the following capabilities:

• Data diagnostics

  This capability helps you analyze the distribution of features in user feature tables and item feature tables, and calculate the conversion rate, retention rate, and recurrence rate based on data in user behavior tables. With this capability, you can check the logs in recommendation scenarios to facilitate the feature and parameter configurations in recommendation solution customization.

• Recommendation solution customization

  This capability allows you to configure various recall algorithms, offline statistical and real-time statistical features, coarse ranking models, and fine ranking models. Then, this capability generates the code of feature engineering, recall models, coarse ranking models, fine ranking models, and the recommendation engine.

• Quick deployment

  You can deploy the code and scripts generated in recommendation solution customization to DataWorks with a few clicks. A deployment flowchart is provided to help you complete feature and training sample data in sequence.

Billable item 2: the recommendation solution implementation and deployment of advanced features

If you want Alibaba Cloud engineers to build a recommendation system and customize solutions, contact Alibaba Cloud customer service. After the custom services are complete, you need to pay for the customization.

Billable item 3: cloud resource consumption

To build a complete recommendation system, you also need to use other Alibaba Cloud services such as MaxCompute and Elastic Algorithm Service (EAS) of Machine Learning Platform for AI (PAI). The resource consumption is charged based on the billing rules of the corresponding cloud services. The fees of cloud resource consumption are not included in the fees of the PAI-Rec platform.

The following table describes the fees for building a complete recommendation system (PAI-Rec Advanced Edition), including offline training and online services. The fees are estimated based on daily active users (DAU).