Beyond Isolated Clients: Integrating Graph-Based Embeddings into Event Sequence Models

2026-04-10

AI Summary

This work addresses the limitation of existing self-supervised event sequence models, which often neglect the global structure of user-item interaction graphs and thus fail to fully capture user behavior. To bridge this gap, the study systematically integrates graph embedding with event sequence modeling and introduces three model-agnostic strategies: enhancing event embeddings with structural information, aligning user representations with graph-derived embeddings, and incorporating a structure-aware pretraining task. The research further uncovers the critical role of graph density in guiding strategy selection and moves beyond the conventional isolated client modeling paradigm. Extensive experiments on four real-world datasets from finance and e-commerce domains demonstrate the effectiveness and generalizability of the proposed approach, achieving up to a 2.3% improvement in AUC.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large-scale digital platforms generate billions of timestamped user-item interactions (events) that are crucial for predicting user attributes in, e.g., fraud prevention and recommendations. While self-supervised learning (SSL) effectively models the temporal order of events, it typically overlooks the global structure of the user-item interaction graph. To bridge this gap, we propose three model-agnostic strategies for integrating this structural information into contrastive SSL: enriching event embeddings, aligning client representations with graph embeddings, and adding a structural pretext task. Experiments on four financial and e-commerce datasets demonstrate that our approach consistently improves the accuracy (up to a 2.3% AUC) and reveals that graph density is a key factor in selecting the optimal integration strategy.

Problem

• event sequence modeling
• graph structure
• self-supervised learning
• user-item interaction
• temporal order

Innovation

• graph-based embeddings
• self-supervised learning
• event sequence modeling
• contrastive learning
• user-item interaction graph