Notification Channel Selection Using Response Probability Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing notification systems in online services struggle to predict member responses accurately, leading to decreased engagement and satisfaction due to inappropriate notification frequency, type, and timing, and lack of personalization across different communication channels.
Innovation Solution
Implementing a deep neural network (DNN) to learn personalized Weibull distribution parameters for member features, enabling precise prediction of member responses to notifications across various channels like badge notifications, push-mobile, and emails, using a shared neural network structure to save computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If notifications are sent frequently to increase member engagement, then member engagement improves, but member satisfaction deteriorates due to notification overload and burnout
Solution Approach 1:
The system dynamically changes notification parameters (timing, frequency, channel selection) based on learned member preferences and contextual factors. The DNN model adjusts these parameters to optimize engagement while preventing burnout by avoiding excessive or poorly-timed notifications.
Solution Approach 2:
The system implements feedback loops where member responses to notifications are monitored and fed back into the DNN model to refine future notification decisions. This allows the system to learn from past interactions and adjust notification strategies to maintain engagement without causing burnout.
2Object-affected harmful factors
If personalized notification strategies are implemented to improve member satisfaction, then member satisfaction improves, but device complexity increases due to multiple prediction models
Solution Approach 1:
The system merges multiple prediction functions into a single integrated DNN model that simultaneously predicts member response probability, visit timing, and optimal notification channel. This consolidation reduces system complexity while maintaining personalized prediction capabilities across different notification scenarios.
Solution Approach 2:
The DNN model is designed as a universal prediction system that handles multiple notification types (push notifications, emails, in-app messages) and predicts various outcomes (click-through, visit timing, engagement level) using a single model architecture, thereby reducing the need for separate specialized models.
3Quantity of substance
If multiple notification channels are used to reach members, then member reach improves, but loss of information increases due to difficulty in selecting optimal channel
Solution Approach 1:
The system dynamically selects notification channels based on real-time member context, device availability, and learned preferences. The DNN model adjusts channel selection probabilities dynamically rather than using fixed rules, allowing optimal channel choice that maximizes reach while maintaining information accuracy.
Data Source
AI summary
Methods, systems, and computer programs are presented for predicting a response probability to a sent notification. One method includes an operation for training respective neural networks to obtain a first, second, and third models. The first model generates an embedding based on member information. The second and third model generate parameters for a distribution function. The first model is used to calculate a member embedding when accessing a notification for a member. Further, the method second model calculates a first parameter value, and the third model calculates a second parameter value based on the member embedding. Further, the method determines, a first probability that the member will visit the online service in response to the notification and a second probability that the member will visit without sending the notification. The method further includes determining to send the notification based on the first probability and the second probability.