Dynamic determination of event type using multitask classification model for increasing efficiency of item pipeline

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

• monitoring interactions of a user with an ordering interface on a user client device associated with the user during a shopping event that is categorized by a type, wherein the type describes a purpose of the shopping event;
• responsive to a monitored interaction being an interaction from a set of trigger interactions, determining, by the computer system, a type of shopping event by the computer system applying the monitored interaction and content of a shopping cart of the ordering interface to a type prediction model, wherein the type prediction model is a multitask classification machine learned model comprising shared layers and multiple output branches, each output branch being a classifier configured to determine whether the shopping event corresponds to a respective type from a set of predefined shopping event types, and wherein the type prediction model was trained by:
• accessing a set of training examples including test interactions, test trigger interactions, content from test shopping carts, and associated types of shopping events,
• applying the type prediction model to the set of training examples to generate training outputs from the multiple output branches, wherein each output branch generates a prediction for its respective shopping event type, and
• backpropagating one or more error terms obtained from one or more loss functions to update a set of parameters of the type prediction model, wherein one or more of the error terms are based on a difference between a label applied to each output branch for a test interaction of the set of training examples and the predictions generated by the respective output branches;
• assigning an updated type to be the determined type;
• and
• performing an action based in part on the updated type.

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:
• monitoring interactions of a user with an ordering interface on a user client device associated with the user during a shopping event that is categorized by a type, wherein the type describes a purpose of the shopping event;
• responsive to a monitored interaction being an interaction from a set of trigger interactions, determining, by the computer system, a type of shopping event by the computer system applying the monitored interaction and content of a shopping cart of the ordering interface to a type prediction model, wherein the type prediction model is a multitask classification machine learned model comprising shared layers and multiple output branches, each output branch being a classifier configured to determine whether the shopping event corresponds to a respective type from a set of predefined shopping event types, and wherein the type prediction model was trained by:
• accessing a set of training examples including test interactions, test trigger interactions, content from test shopping carts, and associated types of shopping events,
• applying the type prediction model to the set of training examples to generate training outputs from the multiple output branches, wherein each output branch generates a prediction for its respective shopping event type, and
• backpropagating one or more error terms obtained from one or more loss functions to update a set of parameters of the type prediction model, wherein one or more of the error terms are based on a difference between a label applied to each output branch for a test interaction of the set of training examples and the predictions generated by the respective output branches;
• assigning an updated type to be the determined type;
• and
• performing an action based in part on the updated type.