Rolling Forward: Enhancing LightGCN with Causal Graph Convolution for Credit Bond Recommendation

TL;DR

The paper tackles the challenge of time-evolving user interests and item popularity in finance recommender systems. It introduces a causal graph convolution extension to LightGCN that learns time-specific embeddings using a causal graph built from a sliding window $[t-w,t)$ to ensure forward-looking predictions. Experiments on BNP Paribas credit-bond data show substantial gains in top-k ranking metrics compared to static LightGCN and traditional baselines, particularly with $w=2$ days. The results demonstrate the importance of preserving interaction order and prioritizing recent data in dynamic graphs, while maintaining LightGCN's simplicity and efficiency.

Abstract

Graph Neural Networks have significantly advanced research in recommender systems over the past few years. These methods typically capture global interests using aggregated past interactions and rely on static embeddings of users and items over extended periods of time. While effective in some domains, these methods fall short in many real-world scenarios, especially in finance, where user interests and item popularity evolve rapidly over time. To address these challenges, we introduce a novel extension to Light Graph Convolutional Network (LightGCN) designed to learn temporal node embeddings that capture dynamic interests. Our approach employs causal convolution to maintain a forward-looking model architecture. By preserving the chronological order of user-item interactions and introducing a dynamic update mechanism for embeddings through a sliding window, the proposed model generates well-timed and contextually relevant recommendations. Extensive experiments on a real-world dataset from BNP Paribas demonstrate that our approach significantly enhances the performance of LightGCN while maintaining the simplicity and efficiency of its architecture. Our findings provide new insights into designing graph-based recommender systems in time-sensitive applications, particularly for financial product recommendations.

Figures (3)

• Figure 1: Standard training framework (A), where the full graph is used to predict each batch of events; this approach learns one static embedding per node in the graph and uses these embeddings to learn a static ranking over the full period. Our approach (B) consists of using a rolling window to learn dynamic ranking; the model has to be always forward-looking, and the node embeddings change from one period to another, allowing them to capture the dynamic context.
• Figure 2: Average model performance for various window sizes. Dashed lines denote a conventional algorithm trained using the full dataset. Only 1 layer and ID embeddings are used for GNN models.
• Figure 3: Performance comparison of the main models across different metrics with varying window sizes.