Entity Resolution

The rule-based matching of records using exact or partial key agreement. This foundational capability applies predefined logic to join datasets where identifiers match precisely.

• Exact Matching: Joins records on a shared unique key, such as a tax ID or passport number.
• Rule-Based Matching: Applies cascading logic, e.g., "Match if SSN matches OR (Name AND Date of Birth match)".
• Phonetic Encoding: Uses algorithms like Soundex or Metaphone to index names by pronunciation, catching spelling variations like 'Smith' vs. 'Smyth'.

While fast and explainable, deterministic methods fail when records contain typos, transliteration errors, or deliberate obfuscation.

A statistical approach that computes the likelihood that two records refer to the same entity despite discrepancies. It uses field-level similarity scores weighted by discriminatory power.

• Token-Based Similarity: Algorithms like Jaccard Index and TF-IDF measure overlap between sets of words or n-grams.
• Edit Distance: Levenshtein and Damerau-Levenshtein distances quantify the number of character changes needed to make two strings identical.
• Weighted Scoring: The Fellegi-Sunter model assigns agreement and disagreement weights to each field, producing a composite match probability.

This method is essential for screening against sanctions lists where names are transliterated from non-Latin scripts.

Supervised and unsupervised models that learn complex matching rules from labeled data or latent patterns, outperforming manual rules on dirty, real-world datasets.

• Supervised Classification: Models like XGBoost or neural networks trained on labeled pairs of duplicates and non-duplicates to predict match probability.
• Active Learning: Iteratively queries a human analyst to label the most uncertain pairs, dramatically reducing manual review effort.
• Clustering for Deduplication: Unsupervised algorithms like DBSCAN or hierarchical agglomerative clustering group records into entity clusters without pre-labeled examples.

These models automatically learn that 'Robert' and 'Bob' are equivalent, adapting to cultural naming conventions.

Leverages the connected structure of data to resolve identities through relationships, not just attribute similarity. This is critical for uncovering hidden beneficial owners.

• Entity Relational Graphs: Constructs a network where nodes are entities and edges are shared attributes like phone numbers, addresses, or IP addresses.
• Community Detection: Algorithms like Louvain or Label Propagation identify tightly connected clusters that likely represent a single real-world actor or collusive ring.
• Collective Resolution: A change to one entity's linkage propagates across the graph, resolving entire neighborhoods simultaneously.

A fraudster using multiple synthetic identities but a single shared address will be collapsed into one entity through graph analysis.

The performance optimization layer that prevents the computationally prohibitive comparison of every record against every other record. Blocking reduces the search space by orders of magnitude.

• Standard Blocking: Partitions the dataset into blocks based on a blocking key, such as the first three characters of a surname or a ZIP code. Only records within the same block are compared.
• Sorted Neighborhood: Slides a window over records sorted by a key, comparing only records within the window.
• Canopy Clustering: Uses a cheap, approximate distance metric to create overlapping clusters, then performs expensive comparisons only within each canopy.

Without blocking, resolving a database of 10 million records would require 50 trillion pairwise comparisons.

Real-time entity resolution at the point of transaction or customer interaction, enabling instantaneous risk decisions rather than batch retrospective analysis.

• Streaming Deduplication: Integrates with Apache Kafka or Flink to resolve identities against a master entity index in under 100 milliseconds.
• Incremental Clustering: Updates entity clusters on-the-fly as new transactions arrive, without recomputing the entire database.
• Cache-First Architecture: Maintains a hot cache of active entity profiles in Redis or similar in-memory stores for sub-millisecond lookups.

This capability allows a payment system to recognize that a new account is linked to a previously blocked entity before the transaction is authorized.

The technique of mapping and examining the relationships between entities to identify hidden connections, collusion, and the structural hierarchy of criminal rings. Entity resolution provides the deduplicated, linked node set that makes network analysis possible, transforming fragmented transaction logs and watchlist entries into a coherent graph topology.

• Reveals centrality metrics identifying key orchestrators
• Detects community structures indicative of collusive rings
• Enables link prediction to surface undisclosed relationships

A legal entity with no significant assets or active business operations, often used as a vehicle to obscure beneficial ownership and facilitate illicit financial flows. Entity resolution systems are specifically architected to detect shell corporations by cross-referencing registration addresses, nominee directors, and corporate service provider data to identify clusters of entities sharing identical structural fingerprints.

• Flags mass-registration addresses hosting thousands of entities
• Identifies reused director profiles across unrelated jurisdictions
• Correlates dormant entities with sudden high-value transaction activity