Lighter-X: An Efficient and Plug-and-play Strategy for Graph-based Recommendation through Decoupled Propagation

TL;DR

The paper tackles the scalability barrier of graph-based recommenders by introducing Lighter-X, a plug-and-play framework that compresses both the sparse graph structure and high-dimensional embeddings to achieve parameter complexity $O(hd)$ with $h \ll n$. By replacing dense, node-ID–driven inputs with a low-rank random input $\mathbf{X}$ and training a small $h \times d$ matrix $\mathbf{W}^\prime$, and by decoupling propagation from learning to precompute $\mathbf{Z}$, Lighter-X enables efficient training and inference while preserving the base models' performance. The approach is instantiated as LighterGCN, LighterJGCF, and LighterGCL, and extended to polynomial filters and graph contrastive learning, with extensive experiments showing competitive results and substantial reductions in parameter counts (down to 0.01–1% of baseline) and training time on datasets such as MovieLens, LastFM, Yelp, Alimama, and HuaweiAds. The framework also demonstrates strong performance on non-bipartite and context-aware scenarios, highlighting its broad applicability. Overall, Lighter-X offers a practical, scalable path for deploying high-quality graph-based recommendations in industry-scale settings, where rapid retraining and low resource usage are critical.

Abstract

Graph Neural Networks (GNNs) have demonstrated remarkable effectiveness in recommendation systems. However, conventional graph-based recommenders, such as LightGCN, require maintaining embeddings of size $d$ for each node, resulting in a parameter complexity of $\mathcal{O}(n \times d)$, where $n$ represents the total number of users and items. This scaling pattern poses significant challenges for deployment on large-scale graphs encountered in real-world applications. To address this scalability limitation, we propose \textbf{Lighter-X}, an efficient and modular framework that can be seamlessly integrated with existing GNN-based recommender architectures. Our approach substantially reduces both parameter size and computational complexity while preserving the theoretical guarantees and empirical performance of the base models, thereby enabling practical deployment at scale. Specifically, we analyze the original structure and inherent redundancy in their parameters, identifying opportunities for optimization. Based on this insight, we propose an efficient compression scheme for the sparse adjacency structure and high-dimensional embedding matrices, achieving a parameter complexity of $\mathcal{O}(h \times d)$, where $h \ll n$. Furthermore, the model is optimized through a decoupled framework, reducing computational complexity during the training process and enhancing scalability. Extensive experiments demonstrate that Lighter-X achieves comparable performance to baseline models with significantly fewer parameters. In particular, on large-scale interaction graphs with millions of edges, we are able to attain even better results with only 1\% of the parameter over LightGCN.

Figures (8)

• Figure 1: Performance vs. Training Parameters: Circle sizes represent parameter counts. Baseline models' parameters scale proportionally with embedding size ($d$) and dataset size ($n$), while Lighter-X achieves higher accuracy with more compact parameter sizes.
• Figure 2: Parameter matrix updates during training.
• Figure 3: Percentage of parameter updated more than $k$ times.
• Figure 4: An overview of the proposed Lighter-X framework.
• Figure 5: Comparison of training time per epoch.