Recommendation System Using Historical Future Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is an inherent lag between the training phase and the serving phase of a recommendation system, which leads to degradation in prediction accuracy due to changes in user and content distributions over time.
Innovation Solution
The model serving delay time is calculated to identify historic user and content distribution data that is used to supplement current data for retraining the model, thereby reducing the lag and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the model is trained using current user and content distribution data, then the training is completed efficiently with available data, but the prediction accuracy degrades due to the time lag between training and serving when user and content distributions change
Solution Approach 1:
The system performs preliminary action by identifying and storing historical future data (user distribution data and content distribution data) in advance of when it will be needed for training. By pre-identifying data from the future time point (serving time) rather than using only current data, the system prepares training data that reflects the actual distribution at serving time, thereby eliminating the accuracy degradation caused by temporal shifts between training and serving phases.
2Productivity
If the model training uses only current data, then the training process is simple and fast, but the recommendations become less accurate over time due to distribution shifts
Solution Approach 1:
The system applies dynamics by making the training data selection adaptive and time-aware. Instead of statically using only current data, the system dynamically identifies historical future data based on the relationship between training time and serving time. This dynamic approach allows the training process to adapt to temporal changes in user and content distributions, maintaining high prediction accuracy while preserving training efficiency through automated data selection.
3Measurement precision
If historical future data is incorporated into model training, then the prediction accuracy improves by accounting for distribution changes, but the data processing and model training complexity increases
Solution Approach 1:
The system introduces an intermediary component (the time interval calculation mechanism and historical future data identification module) that bridges the gap between current training data and future serving data. This intermediary automatically calculates the time interval between training and serving, identifies the corresponding historical future data, and prepares it for training. By automating this intermediary process, the system manages the increased data processing complexity while achieving improved prediction accuracy through better-aligned training data.
Data Source
AI summary
One or more computing devices, systems, and/or methods for content recommendations using historical future data are provided. A model serving delay time is computed as an average of training delays of events. A historical data time interval is determined based upon the model serving delay time. A model is trained for predicting user content preferences using historic user distribution data and historic content distribution data associated with the historic data time interval. The model is utilized to generate and provide content recommendations to users.