Corpus Graph Embedding Aggregation for Faster Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Recommender systems face inefficiencies in processing large and evolving corpora of content due to impracticality of comparing embedding vectors of billions of items, necessitating improved methods for generating aggregated embedding vectors that incorporate relevant neighborhood information.
Innovation Solution
A data-efficient graph-based convolutional network system that combines embedding vectors of target nodes with their relevant neighborhoods using recursive neighborhood identification and aggregation techniques, leveraging random walks to determine relevant neighbors and employing MapReduce for efficient processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If embedding vectors of all items in a large corpus are compared to identify relevant content, then recommendation accuracy can be maintained, but processing time and computational resources become impractical for billions of items
Solution Approach 1:
The patent segments the large corpus graph into smaller neighborhood subgraphs around target items. Instead of comparing against all billions of items, the system identifies and processes only the local neighborhood (e.g., items connected within 1-2 hops), dramatically reducing the comparison scope while preserving relevant contextual information for accurate recommendations
Solution Approach 2:
The system pre-computes and stores embedding vectors for all items in the corpus beforehand. This preliminary action allows the recommendation engine to quickly retrieve and compare pre-generated embeddings during query processing, avoiding the need to compute embeddings on-demand and significantly reducing real-time processing time
2Reliability
If embedding vectors of all items in a large corpus are compared to identify relevant content, then recommendation accuracy can be maintained, but computational resources become impractical for billions of items
Solution Approach 1:
The patent segments the large corpus graph into smaller neighborhood subgraphs around target items. Instead of comparing against all billions of items, the system identifies and processes only the local neighborhood (e.g., items connected within 1-2 hops), dramatically reducing the comparison scope while preserving relevant contextual information for accurate recommendations
Solution Approach 2:
The system performs partial action by comparing embedding vectors only against a subset of relevant items (the neighborhood) rather than the entire corpus. This partial comparison approach consumes significantly fewer computational resources while still achieving accurate recommendations through the use of pre-computed embeddings and localized graph traversal
3Reliability
If neighborhood information is incorporated into aggregated embedding vectors to improve recommendations, then recommendation accuracy improves, but processing resources required to identify and combine neighborhood embeddings increase significantly
Solution Approach 1:
The system pre-computes embedding vectors for all items in the corpus beforehand, including their neighborhood information. This preliminary computation of aggregated embeddings eliminates the need to perform expensive neighborhood identification and combination operations during real-time recommendation queries, significantly improving processing efficiency while maintaining accuracy
Solution Approach 2:
The patent segments the neighborhood aggregation process into pre-computation and query phases. During pre-computation, aggregated embeddings incorporating neighborhood information are generated and stored. During query processing, only simple retrieval and comparison operations are needed, separating the computationally intensive aggregation work from the time-sensitive recommendation generation
Data Source
AI summary
Systems and methods for generating embeddings for nodes of a corpus graph are presented. More particularly, operations for generation of an aggregated embedding vector for a target node is efficiently divided among operations on a central processing unit and operations on a graphic processing unit. With regard to a target node within a corpus graph, processing by one or more central processing units (CPUs) is conducted to identify the target node's relevant neighborhood (of nodes) within the corpus graph. This information is prepared and passed to one or more graphic processing units (GPUs) that determines the aggregated embedding vector for the target node according to data of the relevant neighborhood of the target node.