Deep Learning User Segmentation Using Unobserved Behavior Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional targeted marketing systems are limited by their inability to account for unobserved user behaviors, leading to inaccurate user segmentation and prediction, as they only consider behaviors within their view and tracked with digital cookies, missing external influences and behaviors not logged or cookie-deleted.
Innovation Solution
An unobserved behavior segmentation and targeting system that uses deep learning to estimate and incorporate unobserved behaviors, allowing users to belong to multiple segments and perform mixed targeting by generating combined observed-unobserved vectors and embeddings, and utilizing neural networks for prediction and clustering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional targeting systems use only observed behaviors tracked with cookies, then system complexity is reduced, but user segmentation accuracy deteriorates due to missing unobserved behaviors
Solution Approach 1:
The patent introduces an intermediary deep learning model that acts as a mediator between observed behaviors and user segmentation. This model infers unobserved behaviors from observed ones, filling the information gap without requiring direct observation of all user actions. The intermediary layer processes both observed and inferred behaviors to generate accurate user segments, resolving the contradiction between data completeness and system complexity.
Solution Approach 2:
The patent replaces traditional mechanical tracking methods (cookies, direct observation) with a neural network-based inference system. Instead of mechanically tracking every user action, the system uses deep learning models to substitute and infer unobserved behaviors from observed patterns, achieving higher segmentation accuracy while reducing the burden of complete behavioral monitoring.
2Loss of information
If conventional systems track only logged-in users and cookie-based behaviors, then data collection complexity is reduced, but information completeness deteriorates due to missed behaviors outside system view
Solution Approach 1:
The patent applies preliminary action by pre-training deep learning models on observed behaviors before actual segmentation occurs. These pre-trained models capture patterns and relationships that enable the system to infer unobserved behaviors during operation. This preliminary processing of data allows the system to compensate for missing information without requiring complex real-time data collection mechanisms.
Solution Approach 2:
The patent uses copying by creating a virtual representation of user behavior through deep learning models. Instead of directly observing all behaviors, the system creates a simplified model that copies and infers unobserved behaviors from observed patterns. This virtual copy allows the system to access complete behavior information without the complexity of tracking every actual user action.
3Adaptability or versatility
If users are assigned to single segments based on observed behaviors, then segmentation simplicity is maintained, but targeting flexibility deteriorates as users cannot belong to multiple segments
Solution Approach 1:
The patent introduces dynamics by transitioning from static single-segment assignments to dynamic multi-segment memberships. Users are assigned to multiple segments simultaneously based on their inferred behavior patterns, allowing flexible targeting strategies. The system dynamically adjusts segment assignments based on the deep learning model's inference results, enabling a user to be targeted with different content based on which segment they most closely match.
Data Source
AI summary
The present disclosure relates to systems, non-transitory computer-readable media, and methods for incorporating unobserved behaviors when generating user segments or predictions of future user actions. In particular, in one or more embodiments, the disclosed systems utilize a deep learning-based clustering algorithm that segments the behavioral history of users based on a future outcome. Further, the disclosed systems recognize that users may exhibit behaviors that represent two or more segments and allow for targeted marketing to users based on the user’s inclusion in multiple segments.