Graph-Based Machine-Learned Embeddings for Fast Entity Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search algorithms struggle to efficiently surface relevant information from vast datasets within stringent time constraints, particularly in scenarios requiring real-time responses, due to the complexity of managing diverse data types across different hardware and services.
Innovation Solution
A machine-learned model architecture that generates a graph from relational databases, determines contextual representations of data entities, and identifies salient data entities using graph neural networks, enabling rapid retrieval of top k relevant data entities based on user queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing search algorithms are used to surface relevant information from vast datasets, then comprehensive data coverage is achieved, but response time becomes unacceptably long and computational resources are excessively consumed
Solution Approach 1:
The patent pre-generates contextual representations for all data entities in the dataset before queries are submitted. This preliminary processing creates a ready-to-use embedding space where entities are pre-positioned based on their contextual relationships, eliminating the need for computationally intensive real-time analysis during query processing. The graph neural network is trained in advance to establish these contextual representations, enabling rapid retrieval operations.
Solution Approach 2:
The patent replaces traditional mechanical search algorithms with a machine learning-based embedding system. Instead of using conventional keyword matching and indexing mechanisms, the system transforms data entities into vector representations in a continuous embedding space, where semantic relationships are captured geometrically. This substitution enables faster and more accurate retrieval by leveraging vector similarity computations rather than exhaustive text processing.
2Adaptability or versatility
If traditional search algorithms process vast datasets, then all data types across different hardware and services are searched, but the system complexity and computational resource requirements increase significantly
Solution Approach 1:
The patent employs a universal graph neural network model that can process multiple data types through a single unified architecture. The graph structure serves as a universal representation framework that accommodates diverse data entities (files, databases, services, hardware components) within the same system. The embedding model learns shared representations across different data types, enabling the system to handle heterogeneous data without requiring separate processing pipelines for each type.
Solution Approach 2:
The patent transforms complex structured data into continuous vector parameters through the embedding process. By converting discrete data entities into continuous vector representations, the system simplifies the complexity of handling diverse data types. The graph neural network learns optimal parameter transformations during training, mapping complex multi-dimensional data relationships into a lower-dimensional embedding space that preserves semantic relationships while reducing computational complexity.
Data Source
AI summary
Predicting the salience of one or more data entities to a particular (target) data entity from among a plurality of data entities may comprise generating a graph of the plurality of data entities and a machine-learned model architecture that predicts the salience of the one or more data entities output by the machine-learned model architecture using the graph. For example, the machine-learned model architecture may comprise a first machine-learned model for generating an embedding using the content of the target data entity, a second machine-learned model for generating a vector using the data type indicated by the target data entity, and a third machine-learned model (e.g., a graph neural network or other feed-forward neural network) for generating a contextual representation of the target data entity to which other contextual representations associated with the plurality of data entities may be compared (e.g., using Euclidean distance, cosine similarity, dot product).