Boosting House Price Estimations with Multi-Head Gated Attention

TL;DR

The paper addresses the challenge of accurately predicting house prices by capturing complex spatial dependencies. It introduces Multi-Head Gated Attention for spatial interpolation, creating 'house embeddings' by combining Geo and Euclidean attention heads, and demonstrates that these embeddings enable simpler models to match or outperform ensembles. Through experiments on four international datasets, the method shows improved RMSE and MALE over baselines and ASI, with embeddings further boosting baseline models. The work advances spatial interpolation in real estate, offering a robust, data-compressed representation and highlighting potential for multi-modal data integration.

Abstract

Evaluating house prices is crucial for various stakeholders, including homeowners, investors, and policymakers. However, traditional spatial interpolation methods have limitations in capturing the complex spatial relationships that affect property values. To address these challenges, we have developed a new method called Multi-Head Gated Attention for spatial interpolation. Our approach builds upon attention-based interpolation models and incorporates multiple attention heads and gating mechanisms to capture spatial dependencies and contextual information better. Importantly, our model produces embeddings that reduce the dimensionality of the data, enabling simpler models like linear regression to outperform complex ensembling models. We conducted extensive experiments to compare our model with baseline methods and the original attention-based interpolation model. The results show a significant improvement in the accuracy of house price predictions, validating the effectiveness of our approach. This research advances the field of spatial interpolation and provides a robust tool for more precise house price evaluation. Our GitHub repository.contains the data and code for all datasets, which are available for researchers and practitioners interested in replicating or building upon our work.

Figures (3)

• Figure 1: Architecture representation of the multi-head gated-attention-based interpolation. (A) Represent the Euclidean interpolation block based on the multi-head gated Attention. (B) Represent the geo-interpolation block based on the Multi-Head Gated attention.
• Figure 2: Comprehensive overview of the Gated Multi-head Attention mechanism within the Attention Block. (A) depicts the initial computation of geodesic and Euclidean distances, serving as the foundation for subsequent attention calculations. (B) illustrates the Similarity Function, which transforms these foundational distances into similarity scores. (C) shows the core Multi-Head Gated Attention Block, where these similarity scores derive gated attention weights across multiple heads. (D) Highlights the Aggregated Attention Head, consolidating the gated attention weights from all heads into a singular vector. (E) represents the aggregation of multiple gated attentions for each weighted sum. (F) Indicates the Final Attention Vector.
• Figure 3: Analysis of Geodesic and Euclidean Distances Among the 60 Nearest Houses Across Datasets \ref{['fig:subfig_a']} Highlights the variation in quantiles of the average geodesic distance (in km) for the 60 nearest houses across the four datasets, reflecting the spatial proximity of residences. \ref{['fig:subfig_b']} Represents the distribution in quantiles of the average normalised Euclidean distance for the 60 nearest houses, taking into account the structural features of the houses. Min-max normalisation was employed to standardise the distance values due to the diverse attributes of the houses in each dataset.