Shopping Event Type Prediction for Adaptive Item Pipeline Actions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for monitoring user interactions with online systems fail to accurately determine the purpose of interactions based on user intent, limiting their ability to optimize the item pipeline from supplier to user.
Innovation Solution
An online concierge system monitors customer interactions during shopping events, using a type prediction model to dynamically update the type of shopping event based on interactions and shopping cart content, allowing for improved actions such as demand forecasting and targeted advertisements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual observation or basic tracking methods are used to monitor user interactions, then the system is simple to operate, but the ability to accurately determine user intent and purpose is insufficient
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between raw interaction data and intent determination. The model processes user interactions, shopping cart content, and contextual information to infer user intent and shopping event types, resolving the contradiction by providing accurate measurement without requiring complex manual analysis systems
Solution Approach 2:
The patent replaces manual observation and basic tracking methods with an automated machine learning-based system. The model automatically analyzes user interactions and determines shopping event types, substituting mechanical human analysis with computational intelligence to achieve both accuracy and operational simplicity
2Adaptability or versatility
If static event type classification is used, then the system is easier to implement, but the ability to adapt to changing user intent during shopping events is limited
Solution Approach 1:
The patent implements dynamic event type classification by continuously monitoring user interactions during shopping events and updating the predicted event type in real-time. The system transitions from static classification to dynamic reclassification, allowing the system to adapt to changing user intent as the shopping event progresses
Solution Approach 2:
The system incorporates feedback loops where user interactions during shopping events are continuously fed into the machine learning model, which updates the event type prediction based on new information. This feedback mechanism enables the system to adapt to changing user intent while maintaining manageable complexity through automated processing
3Reliability
If basic tracking methods are used, then the system requires less computational resources, but the ability to perform accurate demand forecasting and targeted actions is reduced
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model offline on historical data before deployment. This pre-processing step enables the model to make accurate predictions during runtime with lower computational overhead, resolving the contradiction between forecasting accuracy and real-time resource consumption
Solution Approach 2:
The system dynamically adjusts model inference parameters based on the shopping event stage and available data. By optimizing computational parameters such as model complexity and inference frequency, the system achieves accurate demand forecasting while managing computational resource consumption efficiently
Data Source
AI summary
An online concierge system dynamically determines types of shopping events. The types may be used in various ways to increase efficiency of an item pipeline. The system may monitor interactions of a customer with an ordering interface on a customer client device associated with the customer. The monitoring may be 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, the system may determine a type of shopping event by applying the monitored interaction and content of a shopping cart of the ordering interface to a type prediction model. The system may assign an updated type to be the determined type, and perform an action based in part on the updated type.