Analyzing the effectiveness of large language models in real estate value estimation

Abstract

In this study, we investigate the effectiveness of Large Language Models (LLMs) for the task of estimating transaction prices of real estate properties (used condominiums). The estimation of real estate prices is challenging due to factors such as the lack of sufficient available data for each region, and the prevalence of missing data and spelling inconsistencies. Since LLMs have knowledge about local characteristics and are capable of internally correcting spelling inconsistencies, it is expected that estimation models using LLMs can mitigate these difficulties. Experimental results showed the accuracy of the proposed model using LLMs outperformed other models for properties in regions that were not included in the training data and properties with spelling inconsistencies.