Machine learning model for predicting future availability of items to evaluate potential replacement items during order fulfillment

A computer program product comprising a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by a processor, cause the processor to perform steps comprising:

• receiving an order from a customer at a computer system, the order identifying one or more items and a retailer from which the one or more items are to be obtained;
• identifying an unavailable item in the order;
• obtaining a set of candidate replacement items for the unavailable item;
• generating replacement score for each candidate replacement item of the set of candidate replacement items by applying a replacement model to each candidate replacement item, wherein the replacement model is trained to output a predicted probability of the customer performing a positive action in response to the unavailable item being replaced with the candidate replacement item, the replacement model trained by:
• obtaining a replacement training dataset including a plurality of replacement training examples, each replacement training example including a combination of an item and a replacement item, each training example having a label indicating whether one or more customers previously performed the positive action in response to replacing the item with the replacement item;
• applying the replacement model to each replacement training example of the replacement training dataset to generate a predicted probability of the customer performing the positive action in response to replacing the item with the replacement item;
• scoring the replacement model using a loss function and the label of the training example;
• and
• updating one or more parameters of the replacement model by backpropagation based on the scoring;
• generating a predicted availability for each candidate replacement item of the set of candidate replacement items by applying an availability model applied to each candidate replacement item, wherein the availability model is trained to output a predicted probability of the candidate replacement item being available during a future time period;
• generating a adjusting the replacement score for each candidate replacement item based on a combination of the predicted probability of the customer performing a positive action and the predicted availability of the candidate replacement item;
• wherein the adjusting reduces the replacement score based on a likelihood of the candidate replacement item being unavailable during the future time period;
• selecting one or more candidate replacement items based on the adjusted replacement scores;
• and displaying the selected one or more candidate replacement items to a user of the computer system associated with the order.

A computer system comprising:

• a processor;
• and a non-transitory computer readable storage medium having instructions encoded thereon that, when executed by the processor, cause the processor to perform steps comprising:
• receiving an order from a customer at the computer system, the order identifying one or more items and a retailer from which the one or more items are to be obtained;
• identifying an unavailable item in the order;
• obtaining a set of candidate replacement items for the unavailable item;
• generating replacement score for each candidate replacement item of the set of candidate replacement items by applying a replacement model to each candidate replacement item, wherein the replacement model is trained to output a predicted probability of the customer performing a positive action in response to the unavailable item being replaced with the candidate replacement item, the replacement model trained by:
• obtaining a replacement training dataset including a plurality of replacement training examples, each replacement training example including a combination of an item and a replacement item, each training example having a label indicating whether one or more customers previously performed the positive action in response to replacing the item with the replacement item;
• applying the replacement s-model to each replacement training example of the replacement training dataset to generate a predicted probability of the customer performing the positive action in response to replacing the item with the replacement item;
• scoring the replacement model using a loss function and the label of the training example;
• and updating one or more parameters of the replacement model by backpropagation based on the scoring;
• generating a predicted availability for each candidate replacement item of the set of candidate replacement items by applying an availability model applied to each candidate replacement item, wherein the availability model is trained to output a predicted probability of the candidate replacement item being available during a future time period;
• generating a adjusting the replacement score for each candidate replacement item based on a combination of the predicted probability of the customer performing a positive action and the predicted availability of the candidate replacement item, wherein the adjusting reduces the replacement score based on a likelihood of the candidate replacement item being unavailable during the future time period;
• selecting one or more candidate replacement items based on the adjusted replacement scores;
• and displaying the selected one or more candidate replacement items to a user of the computer system associated with the order.