User Behavior Prediction Training for Delayed Feedback Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The delay in user behavior actions leads to inaccurate labeling of data, causing incorrect classification of positive samples as negative samples, which affects the accuracy of user behavior prediction models.
Innovation Solution
A method and apparatus for training user behavior prediction models that involve setting two labels for each sample, determining a delay status based on these labels, and performing weighted combination of prediction losses to adjust the model parameters, thereby accounting for different delay statuses of samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the latest data generated within a short time are used to update the machine learning model, then the model can be updated frequently, but the prediction accuracy deteriorates due to delayed feedback causing inaccurate labels
Solution Approach 1:
The patent applies preliminary action by introducing a waiting period before finalizing sample labels. Instead of immediately using the latest data for model updates, the system waits for a predetermined time to elapse after user exposure to target objects, ensuring that delayed behaviors are captured before labeling samples as negative. This preliminary waiting action prevents inaccurate labeling and maintains prediction accuracy while still enabling frequent model updates.
2Speed
If the time interval for model updating is shortened, then the model responds faster to new data, but the label accuracy worsens due to unobserved delayed behaviors
Solution Approach 1:
The system performs preliminary action by establishing a waiting period mechanism before labeling samples. When updating the model with new data, the system does not immediately label samples where user behavior is not observed as negative. Instead, it waits for a predetermined time to pass, allowing delayed behaviors to manifest. This ensures label accuracy is maintained while enabling the model to respond quickly to new data streams.
Solution Approach 2:
The patent implements feedback by continuously monitoring user behaviors after exposure to target objects and using this information to adjust sample labels. The system observes whether users perform specific behaviors within a predetermined time window and uses this feedback to determine whether to label samples as positive or negative. This feedback mechanism ensures that label accuracy is maintained even when model updates occur frequently.
3Ease of manufacture
If positive samples with delayed feedback are marked as negative samples, then the labeling process is simplified, but the prediction accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by implementing a waiting period before finalizing sample labels. Instead of immediately labeling samples as negative when no behavior is observed, the system waits for a predetermined time to elapse. This preliminary action maintains labeling simplicity while preventing the misclassification of positive samples, thereby preserving prediction accuracy.
Solution Approach 2:
The system uses feedback to dynamically adjust sample labels based on observed user behaviors. After the predetermined waiting period, if a user performs the expected behavior, the sample is labeled as positive; otherwise, it is labeled as negative. This feedback-based labeling approach maintains simplicity while ensuring accuracy by only making final labeling decisions after the waiting period confirms the absence of delayed behaviors.
Data Source
AI summary
The disclosure provides methods and apparatuses for training a user behavior prediction model. A sample set generated through streaming is obtained, where any sample includes a sample feature, a first label, and a second label. A sample feature of each sample is input into the model, to obtain a prediction result about whether a corresponding user performs specific behavior. Each corresponding prediction loss is determined based on the prediction result and a first label value. A sample category that indicates a delay status and to which each sample belongs is determined based on the first label value and a second label value, and a weight value of each sample is determined based on the sample category. Weighted combination is performed on each prediction loss corresponding to each sample based on the weight value, and a parameter of the model is adjusted based on an obtained combined loss.