Entity Resolution Algorithm

The fundamental dichotomy in resolution logic. Deterministic matching uses exact, rule-based comparisons (e.g., Tax ID must match exactly) to link records with absolute certainty. Probabilistic matching employs statistical models like the Fellegi-Sunter algorithm to calculate the likelihood that two records refer to the same entity despite errors, omissions, or variations. Modern systems use a hybrid approach, applying deterministic rules for high-confidence anchors and probabilistic scoring for ambiguous edge cases.

A performance optimization technique that prevents the computationally prohibitive comparison of every record against every other record. Blocking partitions the dataset into mutually exclusive buckets using a blocking key—such as the first three characters of a supplier name or a phonetic code. Only records within the same block are compared. Sorted neighborhood and canopy clustering are advanced methods that use sliding windows or cheap distance metrics to further reduce the O(n²) comparison space without sacrificing recall.

The process of transforming raw supplier attributes into comparable signals. Key techniques include:

• Phonetic encoding (Soundex, Double Metaphone) to match names that sound alike despite spelling differences
• Token-based similarity (Jaccard, TF-IDF) for comparing multi-word company names
• Edit distance (Levenshtein, Damerau-Levenshtein) to quantify typographical errors in addresses
• Geocoding to normalize physical locations into comparable latitude/longitude pairs

An approach that models records as nodes and their pairwise similarity scores as weighted edges in a knowledge graph. Resolution becomes a community detection or graph partitioning problem. Algorithms like Louvain or label propagation identify clusters of records that all refer to the same real-world entity. This method excels at resolving complex, multi-record linkages where transitive relationships (A matches B, B matches C, therefore A matches C) must be inferred globally rather than through isolated pairwise decisions.

A human-in-the-loop methodology for generating labeled training data when ground truth is scarce. The algorithm identifies records pairs with the highest uncertainty—those near the decision boundary where the model is least confident—and presents them to a human analyst for adjudication. This labeled feedback is then used to retrain the probabilistic model. Active learning dramatically reduces the manual effort required to build a high-performance resolution system by focusing human attention on the most informative examples.

The final stage of the resolution pipeline. Transitive closure ensures that if Record A matches Record B, and Record B matches Record C, then all three are consolidated into a single entity cluster. Canonicalization selects or synthesizes a single 'golden record' that represents the best-known attributes of the entity—such as the most complete address or the most recent legal name. This golden record becomes the authoritative reference for downstream systems like risk scoring engines and procurement platforms.

An analytical method that maps and explores complex corporate ownership structures using graph databases to identify the ultimate individuals who control or profit from a legal entity. Entity resolution is the prerequisite step that:

• Deduplicates nodes representing the same legal entity across different data sources
• Links edges correctly by resolving parent-child relationship conflicts
• Merges attributes like registration numbers and addresses to create a single source of truth

The graph traversal algorithm itself depends entirely on the quality of entity resolution performed during graph construction.

A digitized due diligence framework that automates the collection and verification of a supplier's identity, ownership, and compliance credentials during the onboarding process. Entity resolution serves as the identity backbone by:

• Matching incoming supplier data against existing master records to prevent duplicates
• Cross-referencing tax IDs, registration numbers, and addresses across disparate systems
• Flagging identity conflicts where a single supplier appears under multiple names

A KYS protocol without entity resolution produces fragmented supplier profiles that undermine risk assessment accuracy.

A system that uses AI to map and monitor the network of a supplier's own suppliers, illuminating hidden dependencies and vulnerabilities deep within the extended supply chain. Entity resolution is essential for:

• Normalizing supplier names reported inconsistently across tier-1 disclosures
• Linking sub-tier entities to their correct parent corporations
• Detecting circular relationships where the same entity appears at multiple tiers under different names

Without accurate entity resolution, sub-tier mapping produces a tangled web of duplicate nodes that obscures rather than reveals concentration risk.

An algorithmic process that continuously monitors a supplier's operational and legal posture to identify subtle deviations from agreed-upon regulatory or contractual standards over time. Entity resolution enables drift detection by:

• Maintaining a persistent entity identifier that tracks a supplier across name changes and restructurings
• Linking new adverse events to the correct resolved entity record
• Preventing false negatives where compliance violations attach to an unresolved duplicate rather than the master record

Drift detection fails when the entity being monitored fragments into multiple unresolved identities.