User Item Embedding Storage Segmentation for Recommendation Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems face challenges in efficiently processing and updating user and item embeddings due to the computational expense of relevance estimation models, particularly when dealing with large volumes of user-item interaction data and the need to ensure that item and user embeddings are of the same version for accurate content recommendations.
Innovation Solution
The system employs a method where user embeddings are stored in non-distributed storage and item embeddings in distributed storage, allowing for efficient retrieval and updating of embeddings during online operations, ensuring that both are of the same version by managing storage processes across different storage types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If item embeddings are stored in distributed storage for fault tolerance and scalability, then system reliability is improved, but the time required to store and update embeddings increases
Solution Approach 1:
The patent segments the embedding storage system into two distinct components: distributed storage for item embeddings (prioritizing reliability) and local storage for user embeddings (prioritizing speed). This segmentation allows each storage type to be optimized for its specific requirements without compromising the other, resolving the contradiction between reliability and storage time.
2Measurement precision
If user embeddings are stored in distributed storage for consistency with item embeddings, then version matching is improved, but the computational overhead and storage time increase
Solution Approach 1:
The patent introduces version identifiers as an intermediary mechanism that enables precise version matching between user and item embeddings without requiring both to be stored in the same distributed storage system. The version identifier acts as a mediator that ensures consistency while allowing the flexibility of different storage architectures.
3Stability of the object's composition
If both user and item embeddings are stored in the same distributed storage, then data consistency is improved, but the processing speed and retrieval efficiency decrease
Solution Approach 1:
The patent applies local quality by storing user embeddings locally (with high-speed access characteristics) and item embeddings in distributed storage (with high consistency characteristics). Each storage location is optimized for the specific requirements of its data type, allowing both consistency and speed to be achieved in their respective contexts.
Data Source
AI summary
Methods and servers for storing data associated with users and digital items of a recommendation system having access to non-distributed and distributed storages. The server trains a model based for generating first user and item embeddings. The server stores (i) the first user embeddings in the non-distributed storage, and (ii) the first item embeddings in the distributed storage. The server re-trains the model for generating second user and item embeddings. The server stores (i) the second user embeddings in the non-distributed storage in addition to the first user embeddings, and (ii) second item embeddings in the distributed storage instead of the respective first item embeddings by replacing the respective first item embeddings. When the second item embeddings are stored on each node of the distributed storage, the server removes the first user embeddings associated with the first value from the non-distributed storage.