Visual Similarity Scoring Using Human-Generated Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for determining similarity between items rely solely on objective criteria, failing to account for subjective user preferences, which limits the accuracy of suggesting related items.
Innovation Solution
The use of human-generated data to weight and adjust similarity scores based on user interactions and subjective feedback, combining with objective criteria to improve visual similarity determinations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If objective criteria alone are used to determine similarity, then the system is simple and fast to operate, but the accuracy of similarity determination deteriorates because subjective user preferences are not considered
Solution Approach 1:
The patent combines objective criteria (item attributes, categories, authors) with subjective criteria (user feedback, interactions, preferences) to determine similarity. This merging of multiple data sources resolves the contradiction by achieving higher accuracy through integrated evaluation while managing complexity through systematic processing of combined inputs.
Solution Approach 2:
The system incorporates user feedback and interaction data to continuously refine similarity determinations. By collecting and processing subjective user responses about item similarities, the system improves measurement precision over time while the feedback mechanism itself provides a structured approach to managing the complexity of incorporating human preferences.
2Reliability
If only objective criteria are used for similarity determination, then the system is easy to operate, but the relevance of suggested items deteriorates because user-specific preferences are not incorporated
Solution Approach 1:
The system automatically collects and processes user feedback and interaction data without requiring explicit user configuration. Users simply interact with items naturally, and the system self-service style gathers this data to improve relevance, maintaining ease of operation while enhancing reliability through automated preference learning.
Solution Approach 2:
The system performs preliminary analysis of user preferences and interactions to pre-compute similarity weights and recommendations. By preparing similarity determinations in advance based on accumulated user data, the system improves relevance while keeping the user interface simple and easy to operate.
Data Source
AI summary
Subjective user-generated data can be utilized to determine visually similar items. Various item descriptors can be determined for a pair of items, which can provide an objective measure of visual aspects of those items, such as how similar those items are in color, style, material, or texture. The ways in which users interact with data for those items, either explicitly or implicitly through user behavior, can provide a level of perceived visual similarity on behalf of these users. The perceived, subjective visual similarity data from the users can be used to adjust a weighting of the various item descriptor factors for a pair of items, or otherwise adjust a visual similarity score, such that items selected as being visually similar more accurately reflect the subjective opinions of the users.