Determining item recommendations based on fulfillment centers

A computer-implemented method comprising:

• receiving, from a machine learning model trained using a training dataset that includes historical data, information identifying a plurality of recommended items for selected items of an online process;
• iteratively generating, based on receiving the information identifying the plurality of recommended items from the machine learning model, clusters of:
• a pair of a recommended item of the plurality of recommended items and a center of the recommended item;
• and
• a pair of a selected item of the selected items and a center of the selected item;
• generating embeddings that provide a compact, multi-dimensional representation of the clusters, and
• determining a cluster combination, of cluster combinations based on the clusters and distances between the embeddings;
• and
• transmitting for displaying, on a device at least a subset of the recommended items of the cluster combination wherein the machine learning model is configured to be re-trained based on feedback data associated with the subset of the recommended items of the cluster combination.

A non-transitory computer readable medium storing computing instructions that, when run on a processor, cause the processor to perform operations comprising:

• receiving, from a machine learning model trained using a training dataset that includes historical data, information identifying a plurality of recommended items for selected items of an online process;
• iteratively generating, based on receiving the information identifying the plurality of recommended items from the machine learning model, clusters of:
• a pair of a recommended item of the plurality of recommended items and a center of the recommended item;
• and
• a pair of a selected item of the selected items and a center of the selected item,
• generating embeddings that provide a compact, multi-dimensional representation of the clusters, and
• determining a cluster combination, of cluster combinations based on the clusters and distances between the embeddings;
• and
• transmitting for displaying, on a device, at least a subset of the recommended items of the cluster combination, wherein the machine learning model is configured to be re-trained based on feedback data associated with the subset of the recommended items of the cluster combination.