Graph Query Platform Vector Embedding Search Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query platforms for graph data structures struggle to effectively incorporate both topology and taxonomy when determining search results, leading to less relevant outcomes.
Innovation Solution
A method involving a graph query platform that processes natural language queries by generating vector embeddings, determining similar node property-level embeddings, and expanding search results by identifying similar entities based on graph embeddings and weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a graph query platform uses traditional query methods without vector embeddings, then the system complexity is lower, but the relevance of search results deteriorates
Solution Approach 1:
The patent introduces vector embeddings as an intermediary between the query and the graph data. The vector embedding module converts natural language queries into vector representations, which then serve as mediators to match against graph node properties and expand search results. This intermediary layer enables semantic similarity matching without directly complexifying the core graph query processing.
Solution Approach 2:
The system segments the query processing into distinct modules: a vector embedding module for converting queries to vectors, a similarity determination module for comparing vectors against graph data, and a list expansion module for generating results. This segmentation allows each component to handle specific complexity independently, improving overall relevance while managing system complexity through modular architecture.
2Measurement precision
If the platform only uses topology for search queries, then the query processing is simpler, but the search result relevance deteriorates
Solution Approach 1:
The patent merges topology-based search (graph structure traversal) with taxonomy-based search (property attribute matching) by combining vector embeddings of both types. The list expansion module integrates results from topology queries and taxonomy queries, weighting them according to relevance. This merging enables the system to leverage both structural and attribute-based information to improve search result relevance without requiring complete redesign of existing query processing.
Solution Approach 2:
The system creates a composite search approach by combining multiple data representations (topology vectors and taxonomy property vectors) into a unified similarity assessment. The vector embedding module generates composite representations that capture both structural relationships and attribute characteristics, allowing the search system to evaluate relevance based on multiple dimensions simultaneously.
3Measurement precision
If the platform incorporates both topology and taxonomy using vector embeddings, then the search result relevance improves, but the processing time increases
Solution Approach 1:
The system performs preliminary actions by pre-computing and storing vector embeddings for graph node properties and taxonomy attributes before actual queries are received. The vector embedding module can be pre-configured with embedding models, and similarity thresholds can be pre-established. This preliminary preparation reduces the computational burden during query processing, as the system only needs to compare pre-computed vectors against query vectors rather than processing raw data in real-time.
Solution Approach 2:
The list expansion module implements partial action by selectively expanding search results based on predefined criteria and weightings. Instead of exhaustively processing all possible graph traversals and property combinations, the system applies partial expansion using vector similarity thresholds and top-k results. This partial approach maintains high relevance for most queries while significantly reducing processing time for edge cases requiring deeper analysis.
Data Source
AI summary
In some aspects, the techniques described herein relate to a method including: determining, by an embedding engine, a first plurality of nodes in a graph database; generating, by the embedding engine, a property-level vector embedding for each node of the first plurality of nodes, wherein each property-level vector embedding is based on a node property defined by each node of the first plurality of nodes; determining, by the embedding engine, a second plurality of nodes; generating, by the embedding engine, a node-level vector embedding for each node in the second plurality of nodes, wherein each node-level vector embedding is based on a type of each node in the second plurality of nodes; and persisting, by the embedding engine, each property-level vector embedding and each node-level vector embedding in a vector database with an association to an index key.