Study reveals optimal strategies for selecting training examples to improve LLM-based location prediction accuracy

Researchers compare multiple demonstration selection methods for next POI prediction tasks using large language models

In-context learning (ICL) performance heavily depends on which examples are selected for the model to learn from

Analysis evaluates random selection, embedding-based selection, and task-specific selection alongside simpler approaches like geographical proximity and temporal ordering

Comprehensive evaluation aims to establish best practices for real-world POI prediction applications based on user check-in history