Visual Preference Navigation Across Large Product Image Sets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing e-commerce systems struggle to recommend visually appealing products due to the subjective nature of fashion and other highly personalized areas, as current search technologies rely on keyword-based searches that fail to capture visual desirability and require users to input specific images, leading to inefficient browsing and limited data collection on user preferences.
Innovation Solution
A computer-implemented method that determines user visual preferences by detecting engagement events with item images, refining a visual preference hypothesis through machine-learned feature extraction, and updating the display set with new images to iteratively match user tastes, without relying on aggregate user behavior data.
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 vague or subjective attributes cannot be effectively searched
Solution Approach 1:
The patent replaces keyword-based text search with visual-based search using machine learning models. Instead of relying on text keywords to describe products, the system uses visual feature extraction and comparison to find products matching user preferences, enabling effective search for products with subjective attributes like fashion items.
Solution Approach 2:
The patent introduces visual feature vectors as an intermediary between user queries and product databases. These vectors serve as a bridge that translates subjective visual preferences into computable representations, allowing the system to handle vague attributes by comparing visual features rather than relying on imprecise keywords.
2Measurement precision
If visual similarity-based search is used, then products matching an uploaded image can be found, but users cannot discover new products they haven't seen before
Solution Approach 1:
The patent implements a dynamic recommendation system that adapts to user preferences through iterative interactions. The system starts with visual similarity search but evolves by learning user preferences from engagement events, dynamically adjusting recommendations to balance visual matching with introducing novel products that match inferred preferences.
Solution Approach 2:
The patent incorporates feedback loops where user engagement events (clicks, purchases, time spent) are continuously monitored and used to refine preference hypotheses. This feedback mechanism allows the system to learn from user interactions and improve product discovery by adjusting recommendations based on actual user behavior rather than relying solely on initial visual similarity.
3Adaptability or versatility
If category-based browsing is used, then all existing products can be displayed, but the browsing experience becomes time-consuming and overwhelming
Solution Approach 1:
The patent performs preliminary action by pre-extracting visual features for all products in the database and organizing them in feature space before user interaction. This preprocessing allows the system to quickly retrieve and recommend products based on visual similarity and inferred preferences without requiring users to browse through all products manually, significantly reducing browsing time while maintaining comprehensive product coverage.
4Ease of operation
If random product display is used, then all products have equal visibility, but visually appealing products are infrequent and shopping fatigue sets in
Solution Approach 1:
The patent changes the display parameters from random ordering to preference-based ordering. By adjusting the ranking parameters according to inferred user preferences and visual appeal metrics, the system prioritizes displaying visually appealing products first while maintaining accessibility to all products through pagination and filtering, thereby reducing shopping fatigue and improving shopping efficiency.
Data Source
AI summary
Determining a visual preference of a user by selecting a display set of item images from at least one item database and causing the item images of the display set to be displayed at a user interface. Engagement events at the user interface between the user and engaged-with item images are detected and used to determine a visual preference hypothesis for the user based on visual features extracted from the engaged-with item images. New item images are selected from the database(s) by comparing their visual features with the visual preference hypothesis. This is an iterative process, in which the visual preference hypothesis is refined and the display set continues to be updated accordingly. In another aspect, an improved user interface facilitates efficient item selection based on active and/or passive engagement events (of various possible types) with an item array, providing a rich source of visual preference information.