Address Harmonization

Institutional collaboration networks are frequently distorted by affiliation variants that refer to the same organization — Wuhan Univ, Wuhan University, department-level address forms — while country-level analyses suffer from inconsistent naming conventions. A two-track module addresses both.

Track 1 — Organization roll-up

1. Each C1 address is parsed into organization candidates.
2. Affiliation variants are clustered by Jaro–Winkler similarity.
3. Each cluster maps to a canonical parent institution (e.g. departments roll up to their university).
4. High-confidence matches consolidate automatically; borderline cases route to review.

Track 2 — Country standardization

Dictionary-based normalization resolves alternative spellings, abbreviations and historical names (USA / United States / U.S.A.; Türkiye / Turkey). Unresolved tokens go to the same review workflow rather than being guessed.

Why it matters

| Without harmonization | With harmonization | | --- | --- | | "Wuhan Univ" and "Wuhan University" are two institutions | One canonical node with combined output | | Country counts split across spelling variants | One row per country | | Institutional rankings undercount fragmented affiliations | Roll-up reflects real institutional volume |

All consolidations are logged and reversible; the canonical mapping is visible per record, so you can always trace an aggregated count back to the original address strings.