Image Search Embedding Retrieval for Long and Obscure Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image search engines use fixed weighting schemes that combine image and landing page features, failing to effectively model 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 exact query matches are absent.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed weighting schemes are used to combine image and landing page features, then the system is simple to implement, but it fails to effectively model semantic relationships and misses relevant results for long or obscure queries
Solution Approach 1:
The patent transforms the fixed weighting parameters into dynamic embedding vectors that are learned through training. Instead of using static weights to combine features, the system uses neural network embeddings that adaptively represent semantic relationships, resolving the contradiction between accuracy and complexity by allowing the system to learn optimal parameter configurations from data
Solution Approach 2:
The patent replaces the mechanical fixed weighting scheme with a neural network-based embedding system. The traditional approach of manually setting weights is substituted with learned embeddings that automatically capture semantic relationships, enabling the system to handle long and obscure queries effectively
2Productivity
If separate indexing systems are used for different languages and query types, then retrieval accuracy for specific query types improves, but system complexity and processing time increase
Solution Approach 1:
The patent creates a universal embedding space that handles multiple languages and query types simultaneously. Instead of maintaining separate indexing systems for different languages and query types, the system uses a single multi-lingual embedding model that can process and retrieve results for any language or query type, thereby improving retrieval speed while maintaining accuracy through the unified semantic representation
3Measurement precision
If term-based retrieval systems are used to identify search results based on key terms, then exact matches are found quickly, but relevant results for long or obscure queries are missed
Solution Approach 1:
The patent introduces embedding vectors as an intermediary between the query and the retrieval process. Instead of directly matching terms, the system first transforms the query into an embedding vector that captures its semantic meaning, then uses this embedding to find relevant results. This intermediary representation allows the system to detect relevance based on meaning rather than exact term matches, improving accuracy for long and obscure queries while maintaining efficient processing through vector similarity search
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.