Beyond Isolated Clients: Integrating Graph-Based Embeddings into Event Sequence Models
Harry Proshian, Nikita Severin, Sergey Nikolenko, Kireev Ivan, Andrey Savchenko + 3 more
TLDR
This paper integrates graph-based embeddings into event sequence models using three strategies to improve prediction accuracy in digital platforms.
Key contributions
- Proposes three model-agnostic strategies to integrate graph structure into contrastive self-supervised learning.
- Strategies include enriching event embeddings, aligning client representations, and a structural pretext task.
- Improves prediction accuracy by up to 2.3% AUC on financial and e-commerce datasets.
- Identifies graph density as a critical factor for selecting the optimal integration strategy.
Why it matters
This work addresses a key limitation in event sequence modeling by incorporating global graph structure. It offers practical, effective methods to boost prediction accuracy for critical tasks like fraud prevention and recommendations. The findings also provide guidance on selecting the best integration approach based on data characteristics.
Original 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.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.