Abstract: A trained model is used to generate action recommendations by predicting different metrics related to items ordered at an online system. The online system gathers replacement data related to replacement of an item either online or at a location of a retailer when a user uses a physical receptacle in communication with the online system. The online system applies a metric prediction model trained to predict, based on the replacement data, an irreplaceability score indicative of an item's irreplaceability feature and/or a brand stickiness score indicative of an item's brand stickiness feature. The online system applies an item replacement model trained to identify, based in part on the irreplaceability score and/or the brand stickiness score, a list of candidate replacement items. A device associated with the user displays a user interface with a replacement item from the list prompting the user to include the replacement item to a cart.

Abstract: A trained model is used to generate action recommendations by predicting different metrics related to items ordered at an online system. The online system gathers replacement data related to replacement of an item either online or at a location of a retailer when a user uses a physical receptacle in communication with the online system. The online system applies a metric prediction model trained to predict, based on the replacement data, an irreplaceability score indicative of an item's irreplaceability feature and/or a brand stickiness score indicative of an item's brand stickiness feature. The online system applies an item replacement model trained to identify, based in part on the irreplaceability score and/or the brand stickiness score, a list of candidate replacement items. A device associated with the user displays a user interface with a replacement item from the list prompting the user to include the replacement item to a cart.