LLM-Based Entity Resolution for Household Movement Detection
A new AI framework uses large language models (LLMs) to detect household movement patterns in mixed-format occupancy data. The approach combines prompt-based named entity recognition, semantic text embeddings, and graph-based reasoning to identify indirect entity links without extensive preprocessing. Evaluated on SPX benchmark datasets (S8-S12), the method addresses challenges such as noise, duplication, and missing identifiers in demographic data.
Key facts
- Entity resolution typically relies on pairwise similarity comparisons
- Household movement involves multiple individuals relocating together across addresses
- Data is mixed-format, noisy, duplicated, and lacks stable identifiers
- Framework integrates prompt-based LLM named entity recognition
- Uses semantic text embeddings for robust similarity computation
- Graph-based reasoning infers group-level movement patterns
- Evaluated on SPX benchmark datasets S8-S12
- No extensive preprocessing required
Entities
—