Amazon Web Services (AWS) recently announced an advertising feature in Personalization to explicitly recommend specific items based on business rules.
Amazon Personalize enables businesses to improve customer retention and monetization metrics by recommending personalized items to customers. By using AWS Personalize, engineers could start developing recommendation engines without prior experience with machine learning and recommendation systems.
Businesses use promotions as one of the main strategies to increase user retention and sales. Promotions are considered part of regular recommendation to meet business objectives. As the blog post says:
You can use promotions to make a percentage of your referrals for any application regardless of domain of a certain type. For example, in e-commerce applications, you can use this feature to mark 20% of recommended items as on sale or from a specific brand or category. For video-on-demand use cases, you can use this feature to fill 40% of a carousel with newly launched shows and movies that you want to highlight, or to promote live content. You can use promotions in domain dataset groups and custom dataset groups (user personalization recipes and similar articles).
The following diagram shows how promotions are used in referral pipelines:
Recommendation with actions
As shown in the figure, the promotional items are defined in the system catalog and then loaded into the input data set. The personalized attempt to recommend the action based on the assigned percentage in the overall recommendation. For example, if we set the count to 50%, about 50% of the recommended items will come from promotional items.
The blog post describes the process of creating an advertising feature simply as follows:
Amazon Personalize makes configuring promotions easy: First, create a filter that selects the items you want to promote. You can use the Amazon Personalize console or API to create a filter with your logic using the Amazon Personalize DSL (Domain Specific Language). It only takes a few minutes. Then, when requesting recommendations, specify the promotion by specifying the filter, the percentage of recommendations that should match that filter and, if necessary, the dynamic filter parameters. The promoted items will be randomly distributed in the recommendations, but existing recommendations will not be removed.
The sample ad application with the detailed steps is shared with the source code on Github.
Having referrals is very important for many online businesses. This motivates many cloud providers such as Amazon, Google and Microsoft to offer recommendations and additional features on their platforms.
