LUNAR Benchmark Evaluates LLM Personalization via Universal User Behavior Logs

LUNAR introduces a benchmark for cross-domain personalization using longitudinal app interaction histories across clothing, food, and mobility. It uses a coarse-to-fine synthesis pipeline to model real-world behavioral distributions for evaluation.