Iterative Visual Search via Embedding Space Geometric Constraints
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer search technologies often fail to accurately satisfy users' search goals, as the results returned from queries, whether in structured query language, natural language, speech, or reference images, do not meet the intended search objectives, requiring users to repeatedly refine their queries.
Innovation Solution
A novel iterative search technique is employed, where a database of documents is embedded into an embedding space, allowing users to interactively refine their search by selecting subsets of documents, which geometrically constrains the search space, presenting a new set of candidate documents closer to the selected subset, and iteratively narrows the search results based on user feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional search technologies are used to return query results, then users can obtain search results quickly, but the relevance and accuracy of the results do not meet user search goals
Solution Approach 1:
The patent implements an iterative search process where users provide feedback by selecting preferred documents from candidate sets. The system uses this feedback to geometrically constrain the embedding space and generate refined candidate sets in subsequent iterations, progressively improving search result relevance based on user preferences
Solution Approach 2:
The search system dynamically adapts the candidate set based on user feedback received in previous iterations. The embedding space is geometrically constrained iteratively, transforming the static search process into a dynamic interaction that evolves toward higher relevance
2Measurement precision
If users repeatedly refine their queries to achieve desired search goals, then search result accuracy improves, but the complexity and time required for the search process increases
Solution Approach 1:
The system performs preliminary actions by pre-computing document embeddings and organizing them in an embedding space before the actual search interaction. This preparation enables rapid geometric constraints and candidate set generation during the iterative refinement process, reducing computational complexity during user interaction
Solution Approach 2:
The patent introduces an intermediary mechanism - the geometric constraint system operating in embedding space - that mediates between user feedback and search result generation. This intermediary translates user preferences into mathematical constraints that automatically refine the search space without requiring users to manually construct complex queries
Data Source
AI summary
Roughly described, a system for user identification of a desired document. A database is provided which identifies a catalog of documents in an embedding space, the database identifying a distance in the embedding space between each pair of documents corresponding to a predetermined measure of dissimilarity between the pair of documents. The system presents an initial collection of the documents toward the user, from an initial candidate space which is part of the embedding space. The system then iteratively refines the candidate space using geometric constraints on the embedding space determined in response to relative feedback by the user. At each iteration the system identifies to the user a subset of documents from the then-current candidate space, based on which the user provides the relative feedback. In an embodiment, these subsets of documents are more discriminative than the average discriminativeness of similar sets of documents in the then-current candidate space.