Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Temporal Graph Network Framework for Dynamic Recommendation

Yejin Kim, Youngbin Lee, Vincent Yuan, Annika Lee, Yongjae Lee

TL;DR

Addressing the challenge of evolving user preferences, this work directly applies Temporal Graph Networks to recommender systems. It introduces memory embeddings for nodes and a graph-embedding mechanism over a continuous-time bipartite graph $\mathcal{G}(T)$, jointly learned with time-aware negative sampling and the Bayesian Personalized Ranking loss $\mathcal{L}_{BPR}$. Across MovieLens and RetailRocket, the TGN-based recommender achieves strong recalls, e.g., Recall@20 of 0.2211 on MovieLens and 0.3610 on RetailRocket, outperforming static, sequential, and prior temporal baselines. The results validate temporally aware graph modeling for recommendations and suggest practical scalability enhancements such as temporal hashing for large-scale deployment.

Abstract

Recommender systems, crucial for user engagement on platforms like e-commerce and streaming services, often lag behind users' evolving preferences due to static data reliance. After Temporal Graph Networks (TGNs) were proposed, various studies have shown that TGN can significantly improve situations where the features of nodes and edges dynamically change over time. However, despite its promising capabilities, it has not been directly applied in recommender systems to date. Our study bridges this gap by directly implementing Temporal Graph Networks (TGN) in recommender systems, a first in this field. Using real-world datasets and a range of graph and history embedding methods, we show TGN's adaptability, confirming its effectiveness in dynamic recommendation scenarios.

A Temporal Graph Network Framework for Dynamic Recommendation

TL;DR

Addressing the challenge of evolving user preferences, this work directly applies Temporal Graph Networks to recommender systems. It introduces memory embeddings for nodes and a graph-embedding mechanism over a continuous-time bipartite graph , jointly learned with time-aware negative sampling and the Bayesian Personalized Ranking loss . Across MovieLens and RetailRocket, the TGN-based recommender achieves strong recalls, e.g., Recall@20 of 0.2211 on MovieLens and 0.3610 on RetailRocket, outperforming static, sequential, and prior temporal baselines. The results validate temporally aware graph modeling for recommendations and suggest practical scalability enhancements such as temporal hashing for large-scale deployment.

Abstract

Recommender systems, crucial for user engagement on platforms like e-commerce and streaming services, often lag behind users' evolving preferences due to static data reliance. After Temporal Graph Networks (TGNs) were proposed, various studies have shown that TGN can significantly improve situations where the features of nodes and edges dynamically change over time. However, despite its promising capabilities, it has not been directly applied in recommender systems to date. Our study bridges this gap by directly implementing Temporal Graph Networks (TGN) in recommender systems, a first in this field. Using real-world datasets and a range of graph and history embedding methods, we show TGN's adaptability, confirming its effectiveness in dynamic recommendation scenarios.
Paper Structure (21 sections, 3 equations, 1 figure, 2 tables)

This paper contains 21 sections, 3 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Definition
  4. Continuous Time Dynamic Graph
  5. Methodology
  6. Temporal Graph Networks
  7. Memory embedding
  8. Graph embedding
  9. Model Optimization
  10. Negative sampling
  11. BPR loss
  12. Experiments
  13. Experiments Settings
  14. Datasets
  15. Baseline
  16. ...and 6 more sections

Figures (1)

  • Figure 1: Overview of TGN framework for dynamic recommendation