Enhancing real estate decision-making: a similarity-based approach for property valuation and recommendation

Although similarity metrics have proven valuable in information retrieval and recommendation systems, their application to real estate has been limited. This study examines 1,014 benchmark land lots in Gwangjin-gu, Seoul, and identifies similar lots using Euclidean distance and cosine similarity. Nine land attributes, including site area, road width, and geographical descriptions are used to compute similarity. Being textual data, geographical descriptions are converted into numerical representations using an embedding model. It was found that the land lots identified as similar by Euclidean distance and cosine similarity are nearly identical, suggesting the effective applicability of both metrics to the real estate industry. In addition, we used industry standards to rigorously evaluate the performance of a similarity-based approach and demonstrated that it outperforms the traditional regression model. Our findings indicate that these metrics can enhance property valuation and recommendation by reducing subjectivity and increasing the efficiency of selecting similar land lots.