Image Search Ranking via Virtual Link Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image search engines face challenges in returning relevant and high-quality images due to the reliance on text association, leading to numerous irrelevant results and images that may not match the search query or be of low quality.
Innovation Solution
A method that calculates link-based ranking scores for images using transitional probabilities generated from content-based similarity metrics, allowing for the prioritization of higher quality images in search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If image search engines rely on text association to return search results, then the quantity of results increases, but the quality and relevance of images decrease
Solution Approach 1:
The patent introduces an intermediary ranking system that mediates between the large set of text-matched images and the user's need for high-quality results. By computing quality scores based on multiple criteria (textual relevance, image properties, source reliability) and applying ranking algorithms, the system filters and orders results to present only the most relevant high-quality images, thus resolving the contradiction between quantity and quality.
2Adaptability or versatility
If image search engines return hundreds of results based on text matching, then coverage of potential matches increases, but user experience deteriorates due to difficulty in locating desired images
Solution Approach 1:
The patent segments the large set of search results into ranked categories based on quality scores. By dividing results into tiers (e.g., high-quality matches, moderate matches, lower-quality matches) and presenting them in ordered lists, the system maintains comprehensive coverage while making it easy for users to quickly identify and access the most relevant images at the top of the results.
Solution Approach 2:
The system performs preliminary ranking and filtering of images before presenting them to users. By pre-computing quality scores and ordering results according to these scores, the system prepares the results in advance so that users receive immediately sorted, high-quality matches without needing to manually search through hundreds of unsorted images.
3Quantity of substance
If image search engines prioritize quantity of results, then more images are returned, but the precision of matching search queries decreases
Solution Approach 1:
The patent implements a feedback mechanism where quality scores are computed based on multiple factors including textual relevance, image properties, and source reliability. This multi-criteria feedback system continuously evaluates and ranks images, ensuring that only those with high precision matches to the query are prioritized in the results, while still maintaining comprehensive coverage through structured ranking.
Data Source
AI summary
A system determines ranking scores for objects based on “virtual” links defined for the objects. A link-based ranking score may then be calculated for the objects based on the virtual links. In one implementation, the virtual links are determined based on a metric of content-based similarity between the objects.