Method and system for substituting items

A processing system, comprising:

• a memory comprising computer-executable instructions;
• and a processor configured to execute the computer-executable instructions and cause the processing system to:
• arrange in a pipeline, a first trained model trained on a first training data set comprising names, descriptions, and categories of a plurality of items, a second trained model trained on a second training data set comprising primary categories items, a third trained model trained on a third training data set comprising acceptable and unacceptable pairings of items, and a fourth trained model trained on a fourth training data set comprising rejection rates of items when offered as substitutes, such that outputs of earlier trained models in the pipeline are provided as inputs to later trained models in the pipeline,
• identify a target item;
• filter, using the first trained model, the plurality of items to obtain at least one similar item to the target item, and output the at least one similar item;
• provide the at least one similar item output from the first trained model as input to the second trained model;
• identify, using the second trained model, a primary category of the target item and a primary category of the at least one similar item, and output the primary category of the target item and the primary category of the at least one similar item;
• provide the at least one similar item output from the first trained model, the primary category of the target item from the second trained model, and the primary category of the at least one similar item from the second trained model, as input to the third trained model,
• identify, using the third trained model, potential substitutes for the target item based on the at least one similar item, the primary category of the target item, and the primary category of the at least one similar item;
• and
• provide the potential substitutes to the fourth trained model;
• and
• rank, using the fourth trained model, the potential substitutes based on a rejection probability of each of the potential substitutes,
• wherein the first trained model, second trained model, third trained model, and fourth trained model are configured to generate output at different temporal intervals.

A computer program product embodied on a computer-readable medium comprising code for performing a method, the method comprising:

• arranging in a pipeline a first trained model trained on a first training data set comprising names, descriptions, and categories of a plurality of items, a second trained model trained on a second training data set comprising primary categories for items, a third trained model trained on a third training data set comprising acceptable and unacceptable pairings of items, and a fourth trained model trained on a fourth training data set comprising rejection rates of items when offered as substitutes, such that outputs of earlier trained models in the pipeline are provided as inputs to later trained models in the pipeline,
• identifying a target item;
• filtering, using the first trained model, the plurality of items to obtain at least one similar item to the target item, and outputting the at least one similar item;
• providing the at least one similar item output from the first trained model as input to the second trained model;
• identifying, using the second trained model, a primary category of the target item and a primary category of the at least one similar item, and outputting the primary category of the target item and the primary category of the at least one similar item;
• providing the at least one similar item output from the first trained model, the primary category of the target item from the second trained model, and the primary category of the at least one similar item from the second trained model, as input to the third trained model;
• identifying, using the third trained model, potential substitutes for the target item based on the at least one similar item, the primary category of the target item, and the primary category of the at least one similar item;
• providing the potential substitutes to the fourth trained model;
• and
• ranking, using the fourth trained model, the potential substitutes based on a rejection probability of each of the potential substitutes,
• wherein the first trained model, second trained model, third trained model and fourth trained model are configured to generate output at different temporal intervals.

A computer-implemented method for training machine learning models for determining a substitute for an item of a plurality of items, the method comprising:

• providing a first set of training data comprising at least one of names, descriptions, or categories of the plurality of items;
• training a first machine learning model using the first set of training data, to identify at least one similar item to each item of the plurality of items;
• providing a second set of training data comprising a primary category for each item of a subset of the plurality of items;
• training a second machine learning model using the second set of training data, to identify a primary category of each of the plurality of items;
• providing a third set of training data comprising pairings of items comprising acceptable pairings and unacceptable pairings;
• training a third machine learning model using the third set of training data, to determine potential substitutes;
• providing a fourth set of training data comprising rejection rates of the plurality of items;
• and
• training a fourth machine learning model using the fourth set of training data, to rank the acceptable pairing of items based on a rejection probability of each of the potential substitutes.