Item Image Selection for Online Shopping Concierge Platforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Online shopping concierge platforms face challenges in efficiently presenting item images to customers, which affects customer interaction and order fulfillment.
Innovation Solution
A method using multiple machine learning models to generate composite scores for different images of an item, selecting the most appropriate image based on these scores, and presenting it to the customer through a graphical user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models are used to generate composite scores for image selection, then image selection accuracy and customer engagement are improved, but system complexity and computational resources increase
Solution Approach 1:
The patent divides the image selection process into multiple independent machine learning models, each evaluating specific aspects of item images (e.g., quality, relevance, customer preference). These segmented models generate individual scores that are then aggregated into a composite score, allowing the system to achieve high selection accuracy while maintaining modularity and manageability of system complexity.
Solution Approach 2:
The patent combines the outputs of multiple machine learning models into a single composite score for each image. This merging process integrates various evaluation dimensions (image quality, item relevance, customer engagement metrics) into a unified selection criterion, enabling accurate image selection while consolidating the decision-making process into a manageable framework.
2Measurement precision
If multiple machine learning models are trained and processed, then image relevance and quality improve, but computational time and processing power increase
Solution Approach 1:
The patent trains multiple machine learning models in advance during an offline phase, allowing them to be ready for deployment. This preliminary action enables the models to quickly evaluate images during online operations without requiring extensive real-time computational resources, thus reducing processing time while maintaining high image quality assessment accuracy.
Solution Approach 2:
The patent uses multiple pre-trained machine learning models that can be deployed in parallel or sequentially. By having these models ready-made and potentially distributed across different computing nodes, the system can process images efficiently without repeatedly training models during operation, significantly reducing computational time while maintaining accurate quality assessment.
3Adaptability or versatility
If personalized image selection based on customer interactions is implemented, then customer engagement improves, but data processing requirements and system complexity increase
Solution Approach 1:
The patent applies different evaluation criteria and weighting schemes in the composite score calculation based on individual customer interaction patterns and preferences. This local quality approach allows the system to personalize image selection for each customer without requiring a completely separate processing system, achieving adaptability while managing complexity through targeted customization of the existing framework.
Solution Approach 2:
The patent incorporates customer interaction data (clicks, purchases, views) as feedback signals that continuously refine the machine learning models' understanding of customer preferences. This feedback mechanism enables progressive personalization of image selection, improving adaptability over time while using the same core system infrastructure, thus avoiding exponential growth in complexity.
Data Source
AI summary
An online shopping concierge platform receives data indicating one or more customer interactions associated with a particular item offered by the online shopping concierge platform; identifies a plurality of different and distinct images of the particular item; generates, based at least in part on multiple different and distinct machine learning (ML) models and for each image of the plurality of different and distinct images, a composite score for the image; selects, based at least in part on its respective composite score, an image of the particular item to be presented to the customer; generates data describing a graphical user interface (GUI) comprising a listing of the particular item including the selected image; and communicates to a computing device associated with the customer the data describing the GUI such that the computing device associated with the customer renders and displays the listing.