Database Item Selection Through Iterative Visual Preference Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation engines rely heavily on extracting common behavioral patterns from user data to link item records, which is inefficient and unreliable, especially for items with complex visual appearances.
Innovation Solution
A computer-implemented method that determines a user's visual preference by selecting a display set of item images, detecting engagement events, and updating the display set based on visual features extracted from engaged-with item images, without relying on aggregate user behavior data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional recommendation engines use aggregate user behavior data to train machine learning algorithms, then item records can be linked based on common behavior patterns, but large volumes of user data are required and the approach is ineffective for items with complex visual appearances
Solution Approach 1:
The patent replaces the conventional machine learning-based recommendation engine with a visual feature extraction and comparison system. Instead of using aggregate user behavior data and complex training algorithms, the system extracts visual features directly from item images and compares them to determine recommendations, substituting behavioral analysis with direct visual analysis
Solution Approach 2:
The patent extracts visual features from item images using computer vision techniques, separating the visual appearance analysis from the behavioral pattern analysis. This extraction allows the system to focus on the visual characteristics of items themselves rather than inferring preferences from user behavior data
2Reliability
If recommendation engines rely on aggregate user behavior patterns, then items can be recommended based on historical data, but the system requires significant time to accumulate sufficient data and perform complex behavior pattern recognition
Solution Approach 1:
The patent performs preliminary extraction of visual features from item images and stores them in advance. When a recommendation is needed, the system can immediately compare these pre-extracted visual features without requiring time to accumulate user data or train models, providing instant recommendations based on visual similarity
3Measurement precision
If search functions require users to specify item attributes, then precise searches can be performed, but users must already have specific attributes in mind which limits discoverability
Solution Approach 1:
The patent enables the system to automatically perform the search and recommendation function without requiring users to specify attributes. The visual feature extraction and comparison system autonomously identifies and recommends items based on visual similarity, making the system self-sufficient and eliminating the need for user input
Data Source
AI summary
In one aspect, visual preference of a user is determined 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.