The original title is: "Cross-Lingual Entity Resolution: Why Your Knowledge Graph Has Four Samsungs"

Short summary

A technical deep dive into cross-lingual entity resolution for knowledge graphs, where a single company like Samsung Electronics appears as four different nodes across Korean, English, Japanese, and Chinese scripts. The article presents a pipeline using transliteration to canonical forms, blocking keys to reduce candidate pairs, and Splink's EM estimator with per-language-pair similarity thresholds. Code examples demonstrate practical implementation for multilingual corporate data.

• •Cross-lingual entity resolution requires transliteration to canonical forms before similarity matching
• •Blocking keys reduce 10B candidate pairs to manageable sets by grouping likely duplicates
• •Per-language-pair similarity thresholds (0.75-0.85) account for different transliteration overlap quality

Generated with AI, which can make mistakes.