Efficient Recommendations via Graph Coarsening and Label Propagation

2026-07-24 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, quick

Summary

A new two-stage diffusion framework addresses scalability and efficiency challenges in massive graph-based recommendation systems, particularly in the telecommunications domain. This approach combines graph coarsening with multi-step label propagation. Initially, domain-specific heuristics aggregate nodes into meaningful communities, reducing graph size while preserving business-relevant relationships. An initial diffusion process, using either a Label Propagation Algorithm (LPA) or a Graph Neural Network (GNN), then propagates labels across the coarsened graph for coarse-grained predictions. Finally, a second LPA within subgraphs generates individual user recommendations. On a real-world telecommunications dataset, using LPA in both stages achieved up to +24% NDCG@5 over a full-graph LPA baseline. Incorporating a lightweight GNN in the first stage further boosted NDCG@5 by more than 50%, though it required substantial training and inference time. This coarsening-driven method aims to balance scalability, latency, and recommendation quality.

Key takeaway

For Machine Learning Engineers designing recommendation systems for massive graphs, if you are struggling with scalability and latency, this two-stage diffusion framework provides a compelling solution. You should consider implementing the LPA-only approach to achieve up to +24% NDCG@5 improvement with better efficiency. Alternatively, if your application demands higher accuracy and can tolerate increased training and inference times, explore integrating a lightweight GNN in the initial coarsening stage for over 50% NDCG@5 boost.

Key insights

Graph coarsening combined with multi-step label propagation offers an efficient framework for scalable recommendation systems.

Principles

• Graph coarsening reduces scale while preserving structural representation.
• Domain-specific heuristics can maintain business relevance during node aggregation.

Method

A two-stage diffusion framework first aggregates nodes into communities using heuristics, then applies an initial label propagation (LPA/GNN) on the coarsened graph, and finally uses a second LPA within subgraphs for individual recommendations.

In practice

• LPA-only two-stage diffusion improves NDCG@5 by up to +24% over full-graph LPA.
• Integrating a lightweight GNN in the first stage can boost NDCG@5 by over 50%.

Topics

• Graph Recommendations
• Graph Coarsening
• Label Propagation
• Graph Neural Networks
• Scalability
• Telecommunications

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• F\textsuperscript{2}LP-AP: Fast \& Flexible Label Propagation with Adaptive Propagation Kernel — F²LP-AP offers a training-free, adaptive label propagation method for diverse graph structures.
• RankGraph-2: Lifecycle Co-Design for Billion-Node Graph Learning in Recommendation — Co-designing graph construction, representation learning, and real-time serving stages is critical for scalable billion-node graph retrieval.
• Graph Reduction in Multirelational Networks: A Spreading-Oriented Reduction Benchmark — Graph reduction's impact on influence maximization performance varies significantly by network type and reduction strategy.
• Clustering Node Attributed Networks with Graph Neural Networks and Self Learning — The framework iteratively refines graph clustering by integrating GNN-generated node representations with self-learning and context graph updates.
• Signed-Graph Recommendation as Structural Consistency Maximization — SSC-Loop improves signed-graph recommendation by maximizing structural consistency across multiple layers to overcome noise and sparsity.
• Two-Stage Learned Decomposition for Scalable Routing on Multigraphs — NEPF decomposes multigraph VRPs into node permutation and edge selection for scalable, high-quality routing.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.