Inferred Embedding Vector Generation for Unmapped Items
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content embedding technologies face challenges in generating meaningful embedding vectors for unmapped items across different content types, limiting automated comparisons and analyses, as well as inferring similarity between items of disparate types.
Innovation Solution
The method involves accessing a document corpus, identifying documents containing the unmapped item, averaging embedding vectors of mapped items within those documents, and optionally weighting by importance or frequency to generate an inferred embedding vector, allowing for projection into a multi-dimensional space for similarity analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If embedding vector generators are type-specific to ensure accurate representation of input data, then the quality and accuracy of embedding vectors for mapped items are improved, but the ability to generate meaningful embedding vectors for unmapped items of different content types deteriorates
Solution Approach 1:
The patent introduces mapping items as intermediaries between unmapped items and the embedding space. Mapped items serve as a bridge that connects different content types to the embedding vector generator, allowing unmapped items to be represented through their association with mapped items in the same document corpus.
Solution Approach 2:
The patent creates inferred embedding vectors for unmapped items by copying and averaging the embedding vectors of mapped items that appear in the same documents. This copying approach allows the system to generate meaningful representations for unmapped items without requiring type-specific generators for each content type.
2Reliability
If embedding vectors are generated only for mapped items using type-specific generators, then the reliability of embedding vectors is improved, but the productivity and scope of automated comparisons across diverse content types deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-identifying mapped items within documents that contain unmapped items. This preliminary identification and averaging of embedding vectors enables the system to quickly generate inferred embedding vectors for unmapped items without requiring complex real-time processing, thus improving productivity while maintaining reliability.
3Manufacturing precision
If traditional embedding vector generators are used for each content type, then the manufacturing precision of embedding vectors for specific types is improved, but the device complexity increases due to needing multiple type-specific generators
Solution Approach 1:
The patent makes the single embedding vector generator universal by allowing it to process mapped items of various content types. Instead of having separate generators for each content type, one generator handles multiple types by processing mapped items that serve as representatives for different content categories, thereby reducing device complexity while maintaining manufacturing precision.
Data Source
AI summary
Systems and methods are presented for inferring an embedding vector of an item of a first type into the embedding space. Upon receiving a first time for which there is no embedding vector, documents of a document corpus that include (co-occurrence) both the received item and other items of the same type are identified. Of those other items that have embedding vectors, those embedding vectors are retrieved and averaged. The resulting averaged embedding vector is established as an inferred embedding vector for the received item.