Customer Relation Detection

Detects customer relationship groups from order shipping-address behavior.

The project classifies shared-address patterns into relation types such as family households, co-living groups, local micro-hubs, building clusters, and business locations.

Use Cases

• customer identity analysis
• fraud / reseller investigation
• promotion abuse monitoring
• household-level segmentation
• network-aware CRM and targeting

Relation Types

• family_core
• co_living
• local_micro_hub
• building_cluster
• business_location

Run Demo

pip install -r requirements.txt
PYTHONPATH=src python src/run_demo.py

Outputs:

artifacts/relation_groups.csv
artifacts/relation_edges.csv
artifacts/graph_summary.csv

Repository Structure

src/customer_relation_detection/   core detection logic
data/sample/                       sample order-address dataset
docs/                              methodology and pipeline design
sql/                               sanitized schema template
notebooks/                         demo notebook
artifacts/                         generated outputs

Documentation

• Problem Context
• Relation Taxonomy
• Methodology
• Pipeline Design

Design Notes

The original production approach is SQL-first and incremental: bulk address processing is handled in the warehouse, while Python handles bounded fuzzy-name validation and relation scoring.