Internalizing Address Refinement Service - Episode 4 (A Sad Premonition)

Kurly has significantly reduced its dependence on external vendor APIs by internalizing the standalone building address refinement process using its own building DB. Of the approximately 10.8 million buildings nationwide, excluding approximately 200,000 complex buildings, the rest were classified as standalone buildings, and Kurly built an Address Refinement 2.0 system that performs refinement using only the basic road name address.

Standalone Building Refinement Logic and AddressSearchParam

Processing standalone buildings begins with converting the input address into an AddressSearchParam object. Each item—city/province, city/county/district, road name, building number, etc.—is split into tokens based on spaces, and administrative district names such as 'Seoul Special City' or 'Gyeonggi-do' are standardized and extracted. In particular, when extracting road names, preprocessing was applied to insert spaces at language boundaries in preparation for cases where names are concatenated without spaces, such as 'Garosu-ro20beon-gil133'. Additionally, to handle compound road name cases where a number follows the road name, such as 'Seoro3-ro', the approach was to call the extractRoadName function twice to separately extract mainRoadName and subRoadName before combining them.

Responding to Administrative District Changes and Data Consistency

To respond to administrative district changes by the Ministry of the Interior and Safety (e.g., Jeollabuk-do -> Jeonbuk State, division of Bucheon-si's city/county/district), logic was applied to minimize discrepancies between the building DB and customer request data. Following the Ministry's guideline that building management numbers do not change, refinement results were unified to the currently valid administrative district code even when requests used past names. Flexible search is supported through the replaceSigunguNameByRule function so that even a query using only 'Bucheon-si' matches the detailed city/county/district, and system stability was ensured through log monitoring on days of large-scale changes.

Results and Future Challenges

While standalone buildings account for 98% of all buildings, considering that complex buildings (such as apartments) actually account for a higher proportion of traffic, the result fell short of the expected 90% reduction. Measurements showed that external vendor call volume was reduced by 60%. This is because the order frequency of standalone buildings, such as detached houses, is lower than that of apartments. Kurly is currently conducting additional work to improve the efficiency of address refinement for complex buildings, which account for the remaining 40% of traffic.