I would like to train AWS personalize based on a bunch of different interaction types, for examples "list view", "detail view", "save", "click". Some interactions clearly mean more than others...for example if a user saves an item he presumably really likes it vs a "list view" he might just be scrolling by it and pause briefly.
Is there any way with AWS personalize to tell it that some interactions should have more weight than others?
You currently cannot specify weights for different event types in Personalize. Although intuitively it seems important to be able to do so, it's not as relevant to the HRNN-based recipes/algorithms in Personalize. Personalize recipes such as user-personalization and personalized-ranking (both of which use HRNN) build sequence models from user sessions that are used to learn each user's interest based on a sequence of events rather than specific event types.
However, the SIMS recipe uses item-item collaborative filtering to make related item recommendations based on co-interactions across all users. Therefore, for SIMS it may be more useful in certain cases to model on a specific event type that captures the behavior you're looking to influence recommendations. For example, if you're looking to implement a "frequently bought together" (cross-selling) use case, modeling on "purchase" events only (ideally where users have purchased multiple items) with SIMS will provide more appropriate recommendations than training on all event types.
In the case where you have a single dataset group with an interactions dataset that has multiple event types and multiple solutions for different use cases (e.g., "recommended for you" on the home page/view using user-personalization trained on all event types and "frequently bought together" on the cart page/view using SIMS trained on "purchase" events only), you can use the eventType field for the CreateSolution API to control whether a single event type or all event types are included in training for each solution.
