Data Quality WorkflowNormalization LogicStructured Address Parsing

Address Normalization System

A data normalization system focused on structured quality improvement.

Research Intern2019Academia Sinica

Address Normalization · Data Quality Pipeline

Data Quality SystemRule-Based ParsingTGIS 2019 Best Paper
  1. 01

    Raw Addresses

    Ambiguous TW Formats

  2. 02

    Token Parse

    City, District, Road & No.

  3. Rule Engine03

    Normalize

    Unified Naming & Digits

  4. 04

    Attribute DB

    PostgreSQL Lookup Match

  5. 05

    Web Sampling

    Django QA Review Flow

  6. 06

    GIS / API Out

    Clean Downstream Data

Combines rule-based normalization, a PostgreSQL attribute database, and Django sampling review for GIS quality.

Django + PostgreSQL · Best Paper
Workflow overview. Some names and data are anonymized or recreated for confidentiality.

Before

Address data across datasets was inconsistent, unstructured, and difficult to analyze reliably.

After

Normalization logic, structured address attributes, and a verification workflow were combined to improve data quality at scale.

Context & Background

  • Research datasets contained large amounts of inconsistent address text.
  • The core need was better downstream data quality, not just string parsing by itself.

Constraints

  • Address formats were highly ambiguous and varied across sources.
  • Users needed a way to inspect and verify normalization results.

Problem Solved

  • Unstructured address formats made downstream analysis inconsistent and error-prone.
  • Normalization required both logic and review support.

My Role

  • Both the parsing logic and the surrounding workflow for checking results were designed together.
  • The goal was to turn messy raw data into something structured enough to reuse confidently.

System Design

  • Combined normalization rules, an address attribute database, and a web-based sampling workflow.
  • Used structured attributes to make downstream handling more consistent.

Implementation Highlights

  • Built rule-based normalization and structured address attributes.
  • Added a web workflow for review and sampling.
  • Framed the work as a data quality system rather than a single parser.

Outcome & Impact

  • Improved data quality for address-heavy datasets.
  • Delivered a more reliable normalization workflow for future GIS and research use.

Tech Stack

PythonDjangoPostgreSQLREST API