GLMix Non-linear Optimization for Communication Targeting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Social networking services face challenges in optimizing the combination of communication types, channels, and volume to effectively engage users, as improper targeting can lead to reduced user engagement and increased complaints.
Innovation Solution
A machine learned model integrating Generalized Linear Mixed Models (GLMix) and Non-Linear Optimization is used to personalize communication targeting and volume control, predicting user engagement and negative reactions to maximize relevant interactions while minimizing complaints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If communication volume is increased to improve user engagement, then more users are reached, but user complaints and unsubscribe actions increase
Solution Approach 1:
The patent applies local quality by customizing communication parameters (type, channel, volume) for each individual user based on their specific characteristics, activities, and engagement history. Instead of a uniform communication strategy, the system tailors the communication mix to each user's preferences and behavior patterns, thereby maximizing engagement while minimizing complaints for each specific user segment.
Solution Approach 2:
The system dynamically adjusts communication parameters (type, channel, volume) based on user responses and engagement metrics. By continuously monitoring user interactions and modifying the communication strategy in response to observed behaviors, the system optimizes the balance between engagement and complaint reduction through parameter adaptation.
2Productivity
If communication targeting is improved to increase engagement, then relevant communications are sent, but system complexity increases
Solution Approach 1:
The patent segments the communication strategy into distinct components (communication types, channels, and volumes) that can be independently optimized and controlled. By dividing the complex targeting problem into manageable segments, the system can apply specific optimization techniques to each component while maintaining overall effectiveness, thereby reducing the perceived complexity of the system.
Solution Approach 2:
The system implements feedback mechanisms that monitor user responses to communications and use this information to refine targeting accuracy. By incorporating real-time feedback from user engagements and complaints, the system continuously improves its targeting effectiveness without requiring increasingly complex manual configuration, as the feedback loop automates the optimization process.
Data Source
AI summary
In an example embodiment, a machine learned model that integrates a generalized linear mixed model (GLMix) non-linear optimization is utilized to jointly perform personalized communications targeting and volume control. The machine learned model may be trained to not only maximize user engagement with a notification generally, such as maximizing the total number of people who view, save, or apply for a job associated with a job listing in the communication, but also trained to maximize an end goal of the notification, such as the total number of people who apply for the job associated with a job listing in the communication.