Image Search Retrieval Using Shared Embeddings for Semantic Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image search engines rely on fixed weighting schemes that combine image and landing page features, failing to effectively capture semantic relationships and often miss relevant results for long or obscure queries.
Innovation Solution
An embedding neural network model processes image-landing page pairs and image search queries in a shared embedding space, allowing for the identification of relevant results based on semantic similarity, even when query terms do not fully match.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a fixed weighting scheme is used to combine image and landing page features, then the system structure is simple and easy to implement, but the system cannot effectively capture semantic relationships and misses relevant results for long or obscure queries
Solution Approach 1:
The patent transforms the fixed weighting scheme into a dynamic embedding-based approach where parameters (weights) are learned from data. The embedding neural network model learns optimal parameter configurations during training, allowing the system to adaptively capture semantic relationships while maintaining computational efficiency through learned parameterization.
Solution Approach 2:
The patent replaces the mechanical fixed weighting system with a neural network-based embedding model. Instead of manually configured fixed weights, the system uses learned embeddings that automatically capture semantic relationships through neural network transformations, substituting rigid mechanical combination with flexible data-driven modeling.
2Reliability
If separate indexing systems are used for different query types, then each query type can be optimized specifically, but the system complexity increases and maintenance becomes difficult
Solution Approach 1:
The patent implements a universal embedding space that handles multiple query types (short queries, long queries, obscure queries) through a single unified model. The embedding neural network is designed to be multi-functional, processing diverse query types and image-landing page pairs through the same embedding generation mechanism, eliminating the need for separate indexing systems while maintaining optimization for each query type.
3Productivity
If conventional term-based retrieval is used, then the system is computationally efficient, but it fails to retrieve relevant results for queries with no exact term matches
Solution Approach 1:
The patent introduces embedding vectors as an intermediary representation between raw query terms and retrieval results. Instead of direct term matching, the system transforms queries and images into embedding space where semantic similarity can be computed efficiently through vector operations, serving as a mediator that preserves both computational efficiency and semantic relevance.
Data Source
AI summary
Methods, systems, and apparatus including computer programs encoded on a computer storage medium, for retrieving image search results using embedding neural network models. In one aspect, an image search query is received. A respective pair numeric embedding for each of a plurality of image-landing page pairs is determined. Each pair numeric embedding is a numeric representation in an embedding space. An image search query embedding neural network processes features of the image search query and generates a query numeric embedding. The query numeric embedding is a numeric representation of the image search query in the same embedding space. A subset of the image-landing page pairs having pair numeric embeddings that are closest to the query numeric embedding of the image search query in the embedding space are identified as first candidate image search results.