Online Prediction Model Caching Intermediate Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Customized recommendation systems face challenges with high storage costs, increased access time, and complex calculations due to large amounts of original feature data, particularly in scenarios with a large number of users and items, leading to high machine costs and response delays.
Innovation Solution
The method involves an online prediction model that caches intermediate user and item feature vectors, allowing direct retrieval and reducing the need for repeated calculations, and an apparatus for publishing this model by storing pre-generated intermediate tensor data and model parameters in a database, enabling efficient recommendations without preprocessing original features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If original feature data is stored in the online prediction model, then prediction accuracy is maintained, but storage space consumption increases
Solution Approach 1:
The patent segments the feature data into two parts: intermediate feature vectors that are pre-calculated and stored in the online prediction model, and original feature data that is stored separately in an offline prediction model. This segmentation allows the online model to use only the necessary intermediate representations for fast prediction, while the original data remains in the offline model for reference during updates.
Solution Approach 2:
The patent performs preliminary calculation of intermediate feature vectors offline before deploying the online prediction model. These intermediate vectors are pre-processed and stored in advance, so that during online prediction, the system only needs to retrieve and use these pre-computed vectors rather than calculating from original features in real-time, thus reducing online storage requirements while maintaining prediction accuracy.
2Reliability
If original feature data is stored in the online prediction model, then complete feature information is available, but access time increases
Solution Approach 1:
The patent separates feature access into two layers: the online prediction model stores and provides fast access to intermediate feature vectors, while the offline prediction model stores the complete original feature data. This segmentation enables the online system to access features quickly without loading the entire original dataset, thus reducing access time while maintaining information completeness through the offline model.
Solution Approach 2:
The system pre-calculates and stores intermediate feature vectors in advance during the offline phase. This preliminary action ensures that when online prediction is needed, the system can immediately access these pre-processed vectors without performing time-consuming calculations on the original feature data, thereby significantly reducing access time.
3Reliability
If original feature data is processed in real-time, then up-to-date predictions are made, but calculation complexity increases
Solution Approach 1:
The patent performs the complex feature processing and intermediate vector calculation in advance during the offline phase. The offline prediction model handles the computationally intensive tasks of processing original feature data and generating intermediate representations. During online prediction, the system only needs to perform simple retrieval and combination of these pre-computed vectors, dramatically reducing real-time calculation complexity while maintaining prediction timeliness.
4Reliability
If original feature data is processed in real-time, then accurate predictions are achieved, but response time increases
Solution Approach 1:
The system pre-calculates intermediate feature vectors offline and stores them for quick retrieval. During online prediction, the system retrieves these pre-computed vectors and combines them with minimal additional processing to generate predictions. This preliminary action eliminates the need for time-consuming real-time calculation of features from scratch, thus reducing response time while maintaining prediction accuracy through the use of the pre-processed intermediate representations.
Data Source
AI summary
Provided is a recommendation method. The method includes: determining a target user identifier and a plurality of candidate item identifiers; acquiring the intermediate user feature vector corresponding to the target user identifier by searching for the target user identifier; acquiring the intermediate item feature vectors corresponding to the plurality of candidate item identifiers by searching for the plurality of candidate item identifiers; outputting degrees of matching of the plurality of candidate item identifiers and the target user identifier by predicting the intermediate user feature vector and the intermediate item feature vectors in the online prediction model; and determining a target item identifier to be recommended to the target user identifier from the plurality of candidate item identifiers.