Graph Deconstruction via Key-Value Stores for Real-Time Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems maintain graph data offline, leading to outdated information and significant computing resource consumption, resulting in delayed updates and obsolete recommendations in recommendation engines.
Innovation Solution
Deconstructing graphs into key-value pairs using a database management approach, enabling online generation and updating of embeddings, allowing downstream systems to receive up-to-date information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If graph data is maintained offline in conventional systems, then computing resource consumption is reduced, but information becomes outdated and recommendations are delayed
Solution Approach 1:
The patent segments the graph data into two parts: a static base graph stored offline and dynamic event triggers stored online. This segmentation allows the system to maintain the reliability of the base graph while improving productivity by quickly processing and incorporating new event triggers without requiring full graph reprocessing.
Solution Approach 2:
The system performs preliminary action by pre-computing embeddings for the base graph nodes and edges offline. This allows downstream systems to have immediate access to pre-processed graph information, improving information freshness and update speed when new events occur, as only incremental updates are needed rather than full recomputation.
2Use of energy by stationary object
If graph data is maintained offline, then computing resources are conserved, but update latency increases
Solution Approach 1:
The system implements periodic action by maintaining a continuous online event trigger store that periodically captures and processes new graph events. This allows the system to conserve computing resources by not continuously processing the entire graph, while still reducing update latency by having an always-active mechanism to capture and propagate new information.
Solution Approach 2:
The patent introduces an intermediary event trigger store that mediates between offline graph data and downstream systems. This intermediary captures new graph events in real-time and facilitates quick propagation of updates without requiring direct access to the large offline graph data, thus reducing update latency while conserving computing resources.
3Speed
If embeddings are generated offline, then processing time is reduced for queries, but new content availability is delayed
Solution Approach 1:
The system performs preliminary action by pre-generating embeddings for the base graph offline, which enables fast query processing. Simultaneously, it maintains an online event trigger store that captures new content events in real-time, ensuring that new content is quickly incorporated into the graph structure and made available to downstream systems without waiting for the next offline embedding generation cycle.
Data Source
AI summary
Embodiments of the disclosed technologies are capable of deconstructing a graph using database management approaches. Embodiments describe receiving an event trigger associated with a first node of a graph. The event trigger includes a node identifier of the first node. The embodiments further describe generating a first query for a neighbor identifier associated with a second node of the graph using the node identifier. The neighbor identifier is queried from a first data store. Embodiments further describe generating a second query for a feature corresponding to the neighbor identifier. The feature is queried from a second data store. Embodiments further describe generating an embedding of the event trigger using the node identifier, the neighbor identifier and the feature.