The paper introduces MARS, a modular multi-agent re-ranking framework for repeat-order food delivery recommendation. Its coarse-to-fine pipeline first predicts cuisine and then ranks vendors. MARS combines LightGCN-based global preference signals, Swing-based local peer evidence, geospatial filtering, and prompt-driven LLM reasoning over behavioral, temporal, and geographic context. The authors evaluate it on two Delivery Hero benchmarks, DHRD-SE and DHRD-SG, against heuristic, sequential, graph-based, and food-delivery-specific baselines. The study emphasizes controlled integration, implementation transparency, and reproducible evaluation protocols for hybrid LLM recommenders.
No heat snapshots are available in the last 24 hours.