AWS Entity Resolution New2023年〜
A data matching service that reconciles and unifies records across different data sources
What It Does
AWS Entity Resolution matches and unifies records of the same entity (customers, products, etc.) scattered across different data sources. Even with name spelling variations or address differences, it identifies matching records using rule-based or machine learning-based matching.
Use Cases
It is used for unifying customer data spread across multiple CRM systems, deduplicating marketing databases, and secure data matching between partner companies using data clean rooms.
Everyday Analogy
Think of it like cross-referencing contact lists. When the same person is registered as 'John Smith,' 'J. Smith,' and 'Johnny Smith' across different lists, it automatically finds and merges them into one.
What Is Entity Resolution?
AWS Entity Resolution is a service for data matching and unification. When organizations have multiple systems and data sources, the same customer or product may be registered in different formats. Entity Resolution matches these records and creates a unified view.
Matching Methods
Entity Resolution offers two matching methods: rule-based and machine learning-based. Rule-based matching lets you define matching rules combining attributes like email addresses and phone numbers. Machine learning-based matching uses AWS pre-trained models that account for spelling variations and partial matches. Integration with data provider services also enables matching against third-party data. For detailed coverage of matching methods, books on Amazon are also available.
Getting Started
Create a schema mapping in the Entity Resolution console to define input data attributes. Create a matching workflow, configure the matching method and rules, and run the job. Results are output to S3, where you can review groups of matched records.
Things to Watch Out For
- Matching accuracy depends heavily on data quality and matching rule design, so validate with test data before finalizing rules
- Pricing is based on the number of records processed, and machine learning-based matching costs more per record than rule-based
