Contrastive User Segmentation With User-User Interaction Scores

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing user segmentation technologies on listing platforms are inefficient, leading to increased computing resource consumption, inaccurate search and recommendation results, and wasteful marketing content delivery due to inadequate user segmentation, resulting in repetitive user queries and irrelevant content dissemination.

Innovation Solution

A neural network trained via contrastive learning augmented by user-user scores is used to generate user embeddings, clustering users into segments based on their interactions with items, and employing a generative model to create segment titles and descriptions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional user segmentation methods are used, then implementation is simple, but segmentation accuracy is low leading to irrelevant content delivery

Engineering Contradiction:
Improveuser segmentation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces user-user scores as an intermediary metric that quantifies similarity between users based on their interactions with the same items. These scores serve as a bridge between raw interaction data and the final segmentation clusters, enabling more accurate and interpretable user grouping while maintaining computational efficiency.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transforms user segmentation from traditional demographic-based parameters to behavior-based parameters through user vectors and user-user scores. By changing the fundamental parameters used for segmentation (from static attributes to dynamic interaction patterns), the system achieves higher accuracy in identifying user preferences and behaviors.

Inventive Principle:
Principle #35Parameter changes

2Reliability

If detailed user segmentation is performed, then content relevance improves, but computing resource consumption increases

Engineering Contradiction:
Improvecontent delivery relevanceVSAvoidcomputing resource consumption
Core Design Contradiction:
ReliabilityVSUse of energy by moving object

Solution Approach 1:

The patent performs preliminary computation of user vectors and user-user scores during off-peak times or in batch processing mode. By pre-computing these similarity metrics and storing them, the system avoids real-time computational overhead during content delivery, thus improving relevance without proportionally increasing ongoing computing resource consumption.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates condensed representations of users (user vectors) that capture essential behavioral patterns in a compact form. These vector copies allow for efficient similarity computation and segmentation without requiring access to the full, detailed interaction history of each user, thereby reducing computational complexity while maintaining segmentation quality.

Inventive Principle:
Principle #26Copying

3Productivity

If user segmentation is improved, then marketing efficiency increases, but data processing requirements increase

Engineering Contradiction:
Improvemarketing efficiencyVSAvoiddata processing volume
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent extracts only the most relevant features from user interaction data to construct user vectors. By selectively extracting key behavioral patterns rather than processing all available data, the system achieves effective user segmentation with reduced data processing volume, thereby improving marketing efficiency without proportionally increasing computational burden.

Inventive Principle:
Principle #2Taking out (Extraction)

4Measurement precision

If clustering is performed on user embeddings, then user segments are identified accurately, but processing time increases

Engineering Contradiction:
Improvesegment identification accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent employs dynamic clustering approaches that can adapt to different data distributions and segmentations. The clustering algorithm dynamically adjusts parameters such as cluster density and separation thresholds based on the characteristics of the user embeddings, enabling accurate segment identification while optimizing processing time through adaptive rather than static computation.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4579533A1User segmentation using contrastive learning augmented by user-user scoring
Publication Date: 2025.07.02 EBAY INC
  • EP4579533A1 patent drawingFigure 1
  • EP4579533A1 patent drawingFigure 2
  • EP4579533A1 patent drawingFigure 3

AI summary

Some aspects relate to technologies for generating user segments for users of a listing platform using a neural network trained using contrastive learning augmented by user-user scores. In accordance with some aspects, user vectors are generated for users of a listing platform based on user data tracked by the listing platform. Additionally, user-user scores are generated for pairs of users based on user interactions with items on the listing platform. A neural network is trained using the user vectors and the user-user scores as training data. Once trained, the neural network is used to generate user embeddings by providing user vectors as input to the neural network. User segmentation is performed by clustering user embeddings into clusters with each cluster providing a user segment.