Affinity-Based Dynamic User Clustering for E-Commerce Personalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing e-commerce solutions fail to effectively personalize digital content for users due to oversimplified algorithms that do not account for varying user preferences and noise in data, leading to poor convergence and low response rates to personalized offers.
Innovation Solution
A system using affinity-based dynamic user clustering, which computes affinity scores for users and offers, builds affinity score distributions, and identifies clusters using a Gaussian Mixture Model to provide personalized offers based on user preferences, thereby increasing click-through rates and conversion rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single distribution/bandit arm is used for all users, then the system is simple to implement, but it cannot account for variation in user preferences and causes poor convergence
Solution Approach 1:
The patent segments users into multiple clusters based on their affinity distributions for different offers. Instead of using a single distribution for all users, the system creates separate bandit arms for each cluster, allowing each cluster to have its own response distribution tracked independently. This segmentation enables the system to account for variation in user preferences while maintaining manageable complexity through automated clustering.
2Adaptability or versatility
If discrete tags/segments are used to group users, then the system can group users by characteristics, but it cannot account for offer-specific user preference differences
Solution Approach 1:
The patent implements dynamic clustering where users are assigned to different clusters based on their affinity distributions for specific offers. The clustering is not static or pre-defined but is computed in real-time based on the actual affinity data. This allows the system to adapt to offer-specific preference patterns, where the same user might be in different clusters for different offers, providing precise offer-specific preference accounting.
3Productivity
If overly coarse-grained algorithms are used to simplify processing, then runtime processing is reduced, but personalized content is still not effectively differentiated
Solution Approach 1:
The patent performs affinity score computation and clustering in advance during system initialization or pre-processing phases. By pre-computing the affinity distributions and identifying clusters before runtime, the system reduces the computational burden during actual offer delivery. The clustering structure is established beforehand, allowing runtime operations to simply retrieve and apply the pre-determined cluster assignments without performing complex calculations in real-time.
Data Source
AI summary
The disclosure relates in some cases to a technology for selecting one or more promotions to be presented to online customers using Bayesian bandits and affinity-based dynamic user clustering In some embodiments, a computer-implemented method determines a set of offers is determined, and computes affinity scores measuring affinities of users to items included in the offers. The method builds an affinity score distribution for the offers and identifies clusters of affinity scores for the offers using the corresponding affinity score distribution.