Answers are grounded in this paper and cite their sources.
This article explores the potential of automated valuation models (AVMs) by incorporating features extracted by large language models (LLMs). Utilizing a dataset containing 9,842 residential housing transactions from Oslo, Norway, we extract quantitative features from residential property advertisements with GPT models. We integrate these nonstructured features into an XGBoost model alongside structured data from the same advertisements. The findings reveal that LLM-extracted features significantly boost AVM performance by improving root mean squared error (RMSE) and mean absolute percentage error (MAPE) by 24.31 and 15.25%, respectively. Notably, housing standard and parking are the most influential features. The more accurate property valuations by enhanced AVMs not only have practical implications for tax authorities, banks, real estate agents, and algorithmic traders but also contribute to the research on artificial intelligence.
No takes yet. Share an insight, caveat, or question.
Faxvaag et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: