Dynamic Product Image Grids for Visual Preference Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current e-commerce search technologies struggle to recommend visually appealing products based on subjective user preferences, as they rely on keyword-based searches that fail to capture the inherent visual nature of fashion and other subjective areas, and visual similarity-based searches only suggest similar products without prioritizing desirability, leading to shopping fatigue and limited data collection on user preferences.
Innovation Solution
A system that uses machine learning to extract machine-understandable visual parameters from product images, arranges them in a grid-like display, tracks user interactions, and dynamically rearranges products based on preferences, allowing users to discover visually pleasing items without explicit search queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If keyword-based search is used to find products, then products with well-defined attributes can be found efficiently, but products with subjective visual attributes cannot be effectively searched
Solution Approach 1:
The patent replaces the mechanical text-based search system with a visual-based search system using machine learning models. Instead of relying on keywords and text descriptions, the system uses image processing and visual feature extraction to understand and search for products based on their visual characteristics, thereby resolving the limitation of handling subjective visual attributes
Solution Approach 2:
The patent transforms the search parameters from text-based keywords to visual features extracted from images. By changing the fundamental parameters of the search system from linguistic to visual domain, the system can effectively handle subjective attributes like style, pattern, and appearance that cannot be captured by traditional keyword search
2Ease of operation
If visual similarity-based search is used, then products similar to the uploaded image can be found, but products prioritized by visual desirability cannot be discovered
Solution Approach 1:
The patent implements feedback mechanisms where user interactions with displayed products (clicks, views, purchases) are continuously monitored and used to refine the visual desirability model. This feedback loop allows the system to learn from user behavior and improve its ability to prioritize visually desirable products, transforming static visual similarity search into a dynamic preference-learning system
Solution Approach 2:
The patent performs preliminary analysis of visual desirability by pre-processing and analyzing product images to extract visual features and train machine learning models before the actual search occurs. This preliminary action enables the system to pre-rank products by visual appeal, so when users perform searches, they immediately see prioritized results rather than having to manually evaluate multiple similar products
3Quantity of substance
If category-based browsing is used, then all existing products can be displayed, but the display order is random with regard to visual desirability
Solution Approach 1:
The patent transforms the static, fixed category-based browsing into a dynamic system where product display order continuously adapts based on user interactions and visual desirability metrics. The system dynamically reorders products within categories based on real-time feedback, making the browsing experience adaptive and personalized rather than static and generic
4Device complexity
If static product display is used, then products can be shown in a fixed arrangement, but user preferences cannot be learned effectively
Solution Approach 1:
The patent performs preliminary organization of products into categories and basic visual feature extraction before the user interaction phase, which simplifies the initial system structure. Then, during user interaction, the system progressively learns preferences through feedback without requiring complete system reorganization, thus maintaining relative simplicity while enabling preference learning
Data Source
AI summary
An embodiment of the disclosure provides a system for recommending at least one product to a user device. The system is configured to: (a) receive N product descriptions, each product description in the N product descriptions including an image; (b) extract, for each image in the N product descriptions, a plurality of features including machine-understandable visual parameters; (c) arrange a subset of the N product descriptions relative to a surface in a grid-like manner according to the machine-understandable visual parameters; (d) providing, to the user device, the grid-like arrangement of the subset of the N product descriptions; (e) receive, from the user device, signals indicating an interaction relative to the surface; (f) in response to the signals, rearrange the subset of the N product descriptions relative to the surface; and (g) send, to the user device, the rearranged subset of the N product descriptions.