Image Keyword Association via Visual Similarity and Engagement History
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional search systems face challenges in efficiently identifying and presenting relevant content for visual search queries, particularly when the target image lacks associated keywords, and struggle to provide accurate responses to user queries.
Innovation Solution
The system determines related images based on visual similarity and engagement history, derives candidate keywords from queries associated with these images, filters them, and uses these keywords to present the target image as responsive to user queries, thereby enhancing relevance and user experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional search systems use visual search queries without associated keywords, then the search system can process images directly, but the system cannot efficiently identify relevant content or quickly respond to user queries
Solution Approach 1:
The system performs preliminary actions by pre-generating candidate keywords from queries associated with related images before the actual search query is submitted. This allows the system to have keywords ready for matching, improving search efficiency while maintaining visual search capability. The keywords are derived in advance from engagement history data.
Solution Approach 2:
The patent introduces an intermediary mechanism - candidate keywords - that bridge the gap between visual search queries and content identification. These keywords act as mediators that connect the visual input with the search database, enabling efficient content retrieval without requiring direct keyword input from users.
2Measurement precision
If the system derives candidate keywords from queries associated with related images, then the keywords can be relevant to the target image, but the system complexity increases due to multiple processing steps
Solution Approach 1:
The system applies a universal keyword generation process that works across multiple images and query types. The same mechanism of deriving keywords from engagement history with related images is applied universally to different target images, reducing the need for image-specific complex processing while maintaining high keyword relevance.
Solution Approach 2:
The system changes parameters by using engagement history data (a quantitative parameter) to generate keywords. By adjusting the threshold level of engagement required for keyword selection, the system can control both the relevance and the computational complexity, finding an optimal balance between precision and processing requirements.
3Measurement precision
If the system filters candidate keywords according to engagement history thresholds, then the most relevant keywords are selected, but the processing time and computational resources increase
Solution Approach 1:
The system applies partial action by selecting only the top keywords that meet the engagement threshold criteria, rather than processing all possible keywords. This partial selection approach achieves sufficient keyword accuracy without the excessive processing time that would result from evaluating every possible keyword candidate.
4Adaptability or versatility
If the system associates keywords with target images without initial keyword assignment, then sponsored content delivery can be more targeted, but the initial processing burden increases
Solution Approach 1:
The system performs preliminary keyword association during the image upload and processing phase, so that when sponsored content delivery is needed later, the keywords are already in place. This preliminary action shifts the processing burden to an earlier stage, enabling versatile and targeted sponsored content delivery without increasing the complexity during the actual content distribution phase.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for providing content. One of the methods includes receiving a target image at an image platform; determining one or more images related to the target image; determining queries associated with the related images; deriving candidate keywords from the determined queries; filtering the candidate keywords to select one or more keywords; and providing the image to one or more users in response to respective incoming queries based on the one or more keywords.