Visual Aspect Vector Space Projection for Search Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems are unable to handle user search queries that are outside the scope of predefined labels, leading to inefficiencies in searching items from an inventory, as they fail to match textual queries with visual data effectively.
Innovation Solution
A system that projects visual aspects into a vector space, allowing for the measurement of similarities between visual data inputs, enabling accurate search results without requiring exact textual matches by using a bidirectional image system with modules for receiving, analyzing, generating, and projecting visual data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional textual search queries are used with predefined labels, then the system can handle structured search requests, but it fails to match queries outside the scope of predefined labels
Solution Approach 1:
The patent introduces visual embeddings as an intermediary representation that bridges the gap between user queries and item descriptions. Instead of directly matching textual queries against predefined labels, the system converts both queries and item descriptions into visual embedding vectors, which then serve as the basis for similarity comparison. This intermediary representation enables the system to handle queries outside predefined labels while maintaining reliable match accuracy through vector space similarity measurement.
Solution Approach 2:
The patent transforms the search problem from textual parameter matching to visual embedding space parameter comparison. By converting textual queries and item descriptions into embedding vectors with specific dimensional parameters, the system changes the search parameters from discrete label matches to continuous vector similarity measurements. This parameter transformation allows flexible query handling while maintaining accurate matching through distance-based similarity in the embedding space.
2Reliability
If multiple searches are performed to find relevant items, then comprehensive search results can be obtained, but computing resources such as processor cycles, network traffic, memory usage, storage space, and power consumption increase
Solution Approach 1:
The patent performs preliminary action by pre-computing and storing visual embeddings for all items in the inventory during an offline phase. This preliminary computation of embedding vectors allows the online search phase to simply perform efficient vector similarity comparisons rather than conducting multiple sequential searches. The pre-computed embeddings enable comprehensive search results to be obtained rapidly with minimal computing resource consumption during actual query processing.
Solution Approach 2:
The patent creates a compressed vector space copy or representation of the item inventory through embeddings. Instead of performing multiple searches on the full item database, the system works with compact embedding vectors that capture the essential visual characteristics of items. This copying approach into a lower-dimensional vector space enables efficient similarity search while reducing computing resource requirements compared to multiple full-database searches.
3Measurement precision
If exact textual matches are required for search queries, then precise item identification is achieved, but queries outside predefined labels cannot be handled
Solution Approach 1:
The patent substitutes the mechanical textual matching system with a visual embedding-based similarity measurement system. Instead of relying on exact string matching against predefined labels, the system replaces this rigid mechanical approach with a flexible vector space similarity measurement that can handle both exact matches and approximate visual similarities. This substitution maintains precise item identification through accurate similarity measurement while providing adaptability to handle diverse queries outside predefined labels.
Data Source
AI summary
In various example embodiments, a system and method for projecting visual aspects into a vector space are presented. A query that includes visual data is received by the system from a client device. A visual aspect indicated in the visual data is analyzed. One or more symbols that correspond to the analyzed visual aspect is generated by the system. The analyzed visual aspect is projected into a vector space using the one or more symbols. A group of projections are identified, the group of projections being within a predetermined distance from the projected visual aspect in the vector space. An interface that depicts the further visual aspects is generated. The interface is displayed on the client device.