15-20 sources per investigation
Technology

Entity Resolution

By Seme Research Team · Updated May 22, 2026

Definition

Entity Resolution is the process of determining whether different records or data points refer to the same real-world entity. In identity investigation, entity resolution links profiles across platforms — matching a Twitter handle to a LinkedIn profile, connecting an email address to a real name, or associating a phone number with multiple online accounts. This is one of the most challenging aspects of identity investigation because people use different names, handles, and identifiers across platforms. Modern entity resolution uses probabilistic matching algorithms that consider multiple signals: exact matches (same email, phone), fuzzy matches (similar names, partial matches), behavioral signals (writing style, activity patterns), and contextual signals (shared connections, overlapping timelines).

How It Works

Entity resolution uses a multi-signal matching pipeline. Signal 1 — Identifier Matching: exact matches on email addresses, phone numbers, and profile photos (using perceptual hashing). Signal 2 — Name Similarity: fuzzy string matching using Levenshtein distance, Jaro-Winkler similarity, and phonetic algorithms (Soundex, Metaphone) to handle spelling variations. Signal 3 — Behavioral Matching: comparing writing style (vocabulary, sentence structure), posting patterns (time of day, frequency), and content interests. Signal 4 — Network Overlap: checking if two profiles share connections, follow similar accounts, or belong to the same communities. Signal 5 — Contextual Matching: comparing biographical details (location, employer, education) across profiles. All signals are combined in a weighted scoring model to produce a match probability.

Example

Resolving that "@johndoe" on Twitter, "john.doe@company.com" in a press release, "John Doe, PhD" in an academic paper, and "jdoe82" on GitHub all refer to the same person. The system identifies: exact email match between Twitter bio and press release (Signal 1), name similarity 0.92 between Twitter handle and academic paper author (Signal 2), matching writing style between GitHub comments and Twitter posts (Signal 3), 15 mutual connections between LinkedIn and Twitter profiles (Signal 4), and consistent location and employer data across all profiles (Signal 5). Combined match probability: 97.3%.

Applications

  • Building unified identity profiles from fragmented online presence
  • Anti-fraud detection linking multiple fraudulent accounts
  • KYC compliance connecting customer records across systems
  • Investigative journalism connecting anonymous sources to real identities

Key Statistics

MetricValueSource
Matching Signals5 categoriesSeme methodology
Match Confidence Threshold≥85%Seme configuration
Average Aliases Found3-10 per subjectSeme analytics
Cross-Platform Accuracy94%+Internal benchmarks

Related Terms

Platform Data

15-20
Sources/Investigation
78%
Avg Trust Score
25+
Glossary Terms
10-30 min
Investigation Time

Related Resources