[4F3-GS-10n] AI応用：建設・不動産

[4F3-GS-10n-03] Record linkage of real estate advertising information to improve comprehensiveness of a real estate transaction database

Hiroki Baba³, Tomoko Sekiguchi^2,1, 〇Yoji Kiyota^2,1, Chihiro Shimizu¹ (1. The University of Tokyo, 2. LIFULL Co., Ltd., 3. Kyoto University)

Keywords:real estate, record linkage, clustering, building linkage, dwelling linkage

Real estate information database, an influential data source for comprehensive understanding of real estate transactions, has some problems, since the database is created by multiple real estate intermediary agents. Particularly, we confirm that there are substantial duplication in condominiums and apartments, and thus, it is necessary to integrate the duplicate records together. We regard the task as one of record linkage problems, and develop the model integrating the high likelihood of dwellings with the application to existing data handling techniques such as hierarchical clustering. We then validate whether the integrated records by the proposed method achieve practical recall and precision.