Neural Networks Enable Intelligent Entity Matching in Database Systems

Business software systems and supply chain management platforms struggle with a fundamental data challenge: accurately matching entities across distributed databases. Different systems record company names, addresses, and identifiers using inconsistent formatting, abbreviations, and spelling variations. Graph neural networks now enable intelligent matching that learns from examples, dramatically improving accuracy while reducing manual intervention.

The Problem

Enterprise systems maintain multiple databases containing supplier information, customer records, transaction histories, and inventory data. These systems rarely share consistent data formats or naming conventions. A single supplier might be recorded as “ABC Manufacturing Inc.”, “ABC Mfg”, “American Business Corp”, and variations due to input errors or acquisitions. Without accurate entity matching, systems cannot consolidate records, create reliable reporting, or perform effective supplier relationship management.

Manual matching requires expensive human review and creates bottlenecks in data integration projects. As databases grow to millions of records, manual matching becomes economically impossible. Existing rule-based matching systems require extensive configuration and maintenance, failing when business practices or naming conventions change. The ability to learn from training examples would dramatically improve accuracy while reducing maintenance burden.

What This Invention Does

This patent introduces a graph neural network architecture trained to identify matching entities across databases. The system learns patterns from labeled training examples where humans have already identified matching records. The neural network captures complex relationships between entity attributes including name variations, address formatting differences, phonetic similarities, and historical relationships.

Unlike traditional rule-based systems requiring explicit configuration, the neural network discovers matching patterns automatically through training. Once trained, the system generalizes these patterns to new records it has never seen. The graph-based approach leverages relationships between entities – if supplier A matches supplier B, and B is related to company C, the network can infer probabilistic matches involving C. This relational reasoning enables accurate matching even with sparse or incomplete data.

Key Features

• Graph Neural Network Architecture. Relationships between entities enable intelligent inference beyond simple attribute comparison.
• Training-Based Learning. The system learns from examples rather than requiring manual configuration of matching rules.
• Multi-Attribute Matching. Simultaneously considers names, addresses, contact information, and other attributes with learned weighting.
• Probabilistic Scoring. Returns confidence scores indicating match likelihood, enabling human reviewers to focus on borderline cases.
• Scalability to Large Datasets. Efficiently processes millions of records without proportional increase in computational resources.
• Generalization Capability. Once trained, the system accurately matches new records with attributes and formatting not present in training data.

Who Is Behind It?

Intuit Inc., a leading business software company based in the United States, developed this innovation with a team of machine learning experts: Malathy Muthu, Samvid Jhaveri, and Goutham Kallepalli. Their expertise in financial software, data integration, and artificial intelligence combined to address a critical business software challenge. The filing through Davies Collison Cave Pty Ltd indicates strategic patent protection across major markets including Australia.

Why It Matters

For enterprise software vendors, this technology enables more intelligent data integration features, creating competitive advantages in supply chain management, financial systems, and business intelligence platforms. The ability to automatically match entities without manual configuration reduces customer implementation costs and accelerates time-to-value for data consolidation projects.

For users of enterprise software, accurate entity matching directly improves data quality, enabling reliable financial reporting, effective supplier management, and accurate relationship analysis. The system reduces manual data cleanup work, freeing expensive technical resources for higher-value activities. Better data quality cascades throughout the organization, improving decision-making across all business functions.

The technology enables new use cases for business software including intelligent duplicate detection, cross-system reconciliation, and relationship discovery. The graph-neural-network approach positions the technology at the frontier of machine learning applications in business software, creating barriers to entry for competitors lacking equivalent expertise.

Related Concepts

Entity resolution (also called record linkage or deduplication) is a long-standing challenge in data integration, particularly for enterprise systems where the same real-world entity may appear under dozens of name variations across distributed databases. Traditional rule-based approaches require constant manual maintenance as data formats evolve.

Graph neural networks extend standard deep learning by modelling relationships between data points as graph edges, enabling the network to propagate information across connected entities. This relational structure makes them particularly well-suited to matching problems where context – not just individual attributes – determines identity.

---

AU 2025201913 was published in the Australian Official Journal of Patents on 19 March 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.