Dynamic Push Information Selection Using Thompson Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information processing methods for dynamic push information suffer from low accuracy due to random selection and inefficient creativity selection, leading to wasted exposure opportunities and poor user engagement.
Innovation Solution
An information processing method that counts historical push information feedback data, generates a probability distribution based on click-through rates, and selects target push information for optimal exposure, using Thompson sampling to predict click-through rates and prioritize dynamic push information based on user preferences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If random selection of dynamic push information is used, then implementation of creativity selection is simplified, but accuracy of information processing deteriorates
Solution Approach 1:
The patent introduces feedback mechanisms by counting exposure data and click data from historical push information, generating probability distributions based on this feedback, and using Thompson sampling to continuously optimize selection accuracy while maintaining operational simplicity
Solution Approach 2:
The patent changes the selection parameter from uniform random selection to probability-based selection using Thompson sampling, where selection probability is dynamically adjusted based on historical performance data, thereby improving accuracy without significantly increasing operational complexity
2Adaptability or versatility
If multiple pieces of dynamic push information are pushed simultaneously, then creativity selection can be implemented, but exposure opportunities are wasted
Solution Approach 1:
The patent uses feedback from historical exposure and click data to generate probability distributions that guide selective pushing, ensuring that exposure opportunities are allocated to high-performing creativities rather than being wasted on low-performing ones
Solution Approach 2:
The patent selects a preset quantity of top-performing push information based on probability distributions rather than pushing all available dynamic push information, thereby avoiding waste of exposure opportunities on low-quality content while maintaining creativity selection
3Measurement precision
If probability distribution and Thompson sampling are used, then accuracy of information processing is improved, but computational complexity increases
Solution Approach 1:
The patent transforms the complex problem of optimal selection into a probability distribution estimation problem using Thompson sampling, which provides a computationally efficient approach that balances accuracy and complexity by sampling from beta distributions rather than exhaustive optimization
Data Source
AI summary
In the embodiments of this application, feedback data of historical push information is counted, the feedback data including exposure data and click data. A first probability distribution corresponding to a click-through rate of each piece of push information in the historical push information is generated based on the exposure data and the click data. First predicted click-through rates of to-be-pushed push information are determined according to the first probability distribution, and a preset quantity of pieces of first push information are selected from the to-be-pushed push information according to the first predicted click-through rates. A target predicted click-through rate of each piece of first push information in the first push information is obtained by using a preset target click-through rate prediction model, and target push information is selected for pushing from the first push information according to the target predicted click-through rate.