Community Data Normalization Using Prefix Matching and Text Similarity

Background: Accurate community (小区) information is crucial for rental platforms. Multiple data sources provide heterogeneous, redundant, and noisy records, making it necessary to aggregate and resolve hierarchical relationships among communities.

Goal: Identify and unify duplicate or synonymous community entries, distinguish parent communities from sub‑communities and building addresses, and supplement missing parent or sub‑community data.

Approach: The method leverages the observation that sub‑communities share a longest common prefix with their parent, and that community names follow recognizable patterns (e.g., "PP[数字|字母]区", "PP[数字]幢"). A four‑step pipeline is proposed:

Prefix‑based clustering to build approximate community trees.

Geographic filtering to remove communities whose GPS distance from the tree root exceeds 2 km.

Similarity‑driven re‑clustering using Levenshtein distance and GPS proximity (similarity > 2 and distance < 1 km) to merge synonymous trees.

Final normalization by merging overlapping trees and assigning parent‑child relationships via pattern matching.

Data Pre‑processing: Raw records are standardized by city/district formatting, GPS conversion to Gaode coordinates, and removal of punctuation, retaining only Chinese characters.

Evaluation: Using a manually labeled dataset from Gaode Maps, the algorithm achieves false‑positive rates below 8 % and false‑negative rates below 5 %, demonstrating reliable community normalization.

Conclusion: By exploiting naming conventions and hierarchical patterns, the proposed text‑matching and similarity analysis provides a simple yet accurate solution for community data normalization, enhancing search efficiency and data quality in real‑estate applications.