Machine Learning-Assisted Data Matching for Enterprise Master Data Platforms
Main Article Content
Abstract
Enterprise master data maintenance, encompassing creation, update, retirement, and quality management for the authoritative enterprise view, often requires matching processes for duplicate resolution, governance, and hierarchy management. Such tasks can be achieved using machine learning (ML) models that predict data match quality and support candidate generation and similarity scoring, capable of distributed model execution. An ML-assisted MDM architecture layer is presented, and matching models validated on three distinct business datasets. Metrics show promising performance along with substantial business impact. Duplicate entities across data sources can lead to hindered business processes, poor service quality, and inaccurate reporting. The creation of MDM systems helps mitigate these dangers, but duplicate management through matching is still a challenge. Despite substantial investments in MDM systems, most organizations expend considerable human effort authoring match rules while struggling with duplicate identification across commonly used MDM systems. Duplicates represent a loss of governance and visibility of critical master entity domains. ML assistance can considerably improve candidate generation, similarity scoring, and fusion/merging processes, reducing the manual effort traditionally associated with enterprise MDM duplicate management, enhancing scalability and match quality, and using the MDM data governance framework.