Embedding Space Mapping for Multi-Item Similarity Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Identifying related content items for presentation to a user, especially in response to a query comprising multiple distinct content items, poses significant challenges for online services.
Innovation Solution
A computer-implemented method involving a mapping model that trains to map elements of a single-item embedding vector to a multi-item embedding vector, using a corpus of content groups, and employs Locality Sensitive Hashing to efficiently identify similar content groups based on similarity measures in the multi-item embedding space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional embedding space methods are used to identify related content items, then the system can handle simple queries, but it fails to effectively identify similar content when queries comprise multiple distinct content items of different types
Solution Approach 1:
The patent projects embedding vectors from multiple different embedding spaces (each corresponding to a different content type) into a unified multi-dimensional embedding space. This dimensional transformation enables the system to handle queries comprising multiple distinct content items by representing them in a common space where similarity can be effectively measured, thereby resolving the contradiction between measurement precision and adaptability.
2Reliability
If the system maintains separate embedding spaces for different content types, then it preserves type-specific characteristics, but it increases complexity in identifying similar content across different types
Solution Approach 1:
The patent merges multiple separate embedding spaces into a single unified multi-dimensional embedding space through projection. Each content type maintains its own embedding space with type-specific characteristics, but the projection mechanism combines them into a common space where similarity identification can be performed uniformly, thereby reducing complexity while preserving reliability.
3Ease of operation
If the system uses a unified embedding space for all content types, then it simplifies similarity identification, but it loses the ability to capture type-specific characteristics effectively
Solution Approach 1:
The patent uses dimensional transformation by projecting embeddings from multiple type-specific spaces into a unified multi-dimensional space. This approach maintains the simplicity of unified space operations while preserving type-specific characteristics through the projection process, thereby achieving both ease of operation and measurement precision.
Data Source
AI summary
Systems and methods for identifying content for an input query are presented. A mapping model is trained to map elements of an input query embedding vector for a received query into one or more elements of a destination embedding vector. In response to receiving an input query, an input query embedding vector is generated that projects into an input query embedding space. The input query embedding vector is processed by the mapping model to map the input query embedding vector into one or more elements of a destination embedding vector in a destination embedding space, resulting in a partial destination embedding vector. Items of a corpus of content are projected into the destination embedding space and the partial destination embedding vector is also projected into the destination embedding space. A similarity measure determines the most-similar items to the partial destination embedding vector and at least some of the most-similar items are returned in response to the input query.