Dynamic Threshold Recommendation Engine for Content Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation devices face challenges in simultaneously improving click rates and increasing the number of content distributions, as they either lead to unnecessary distributions that decrease click rates or fail to reach target distribution numbers when focused on high-click probability statuses.
Innovation Solution
A recommendation device that derives expected content distribution values based on user status occurrence probabilities and click likelihoods, extracts combinations with high probability scores, and adjusts a threshold value to ensure both optimal click rates and target distribution numbers are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If content is distributed each time a user status occurs, then the number of distributions increases, but the click rate decreases due to useless distributions
Solution Approach 1:
The patent changes the parameter of threshold value dynamically based on user status and time period. By adjusting the threshold value according to different statuses and time periods, the system identifies content combinations with probability scores meeting or exceeding the threshold, ensuring that distributed content has a guaranteed minimum click probability while still achieving sufficient distribution numbers through systematic parameter adjustment.
2Reliability
If distribution is performed only in predetermined statuses with high click probability, then the click rate improves, but the target number of distributions is not achieved
Solution Approach 1:
The patent makes the threshold value dynamic rather than fixed. The threshold value is adjusted based on the specific user status and time period, allowing the system to expand distribution opportunities to more status-time combinations while maintaining quality control. This dynamic adjustment enables the system to achieve both high click rates and sufficient distribution numbers by adapting thresholds to contextual factors.
3Reliability
If a high threshold value is used to ensure high click rate, then the quality of recommended content improves, but the number of distributions decreases
Solution Approach 1:
The patent segments the distribution strategy by user status and time period, creating different threshold value requirements for different segments. Instead of applying a single uniform threshold, the system divides the problem into multiple status-time segments, each with its own optimized threshold value. This allows high thresholds to be applied where necessary to maintain quality while lower thresholds can be used in other segments to increase distribution volume, achieving both goals simultaneously.
Data Source
AI summary
An information distribution server includes a status estimation unit that derives an expected value of the number of distributions of content in each status in each time period, a score derivation unit that derives a probability score of the content for each time period and for each status, a distribution target determination unit that extracts a combination of a time period, status, and content in which the probability score is equal to or higher than a predetermined threshold value, and determines content relevant to the combination as a distribution target, and a distribution unit that distributes the distribution target on the basis of information indicated by the combination extracted by the distribution target determination unit.