Relation Graph Builder for Visual Item Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content recommendation systems for users without available user data rely on generic category-based recommendations, leading to suboptimal user experience and increased resource consumption, as they fail to account for individual preferences and interests.
Innovation Solution
The system analyzes images to identify associated items and builds a relation graph, using object recognition and segmentation techniques to recommend items based on visual similarity and co-occurrence with the primary content item, even for new or anonymous users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generic category-based recommendations are used for users without available user data, then the system can provide recommendations to new or anonymous users, but the recommendations are not personalized and result in suboptimal user experience
Solution Approach 1:
The system pre-processes and stores visual feature embeddings for items in a relation graph before users arrive. When a user views an item, the system can immediately query pre-computed visual similarities and co-occurrence relationships without waiting for user data accumulation, enabling instant personalized recommendations for new users
Solution Approach 2:
The patent introduces visual feature embeddings as an intermediary representation that bridges the gap between item content and user preferences. Instead of directly matching user profiles (which don't exist for new users), the system uses visual features of viewed items as intermediaries to find and recommend similar items based on visual similarity and co-occurrence patterns
2Device complexity
If generic category-based recommendations are used, then the system can operate without complex processing, but resources are wasted displaying content that may not be of interest to the user
Solution Approach 1:
The patent replaces traditional rule-based category matching (mechanical system) with visual feature embedding and similarity computation. Instead of checking categorical hierarchies, the system uses vector space similarity measurements to identify relevant items, enabling more accurate recommendations without proportionally increasing system complexity
Solution Approach 2:
The system transforms item representations from discrete category labels to continuous visual feature embeddings. This parameter change enables gradient-based similarity computation and allows the system to capture nuanced visual relationships that categorical approaches miss, improving recommendation accuracy while maintaining efficient query performance through pre-computed embeddings
Data Source
AI summary
Recommendations can be generated even in situations where sufficient user information is unavailable for providing personalized recommendations. Instead of generating recommendations for an item based on item type or category, a relation graph can be consulted that enables other items to be recommended that are related to the item in some way, which may be independent of the type or category of item. For example, images of models, celebrities, or everyday people wearing items of clothing, jewelry, handbags, shoes, and other such items can be received and analyzed to recognize those items and cause them to be linked in the relation graph. When generating recommendations or selecting advertisements, the relation graph can be consulted to recommend products that other people have obtained with the item from any of a number of sources, such that the recommendations may be more valuable to the user.