Demo2Vec: Learning Region Embedding with Demographic Information

TL;DR

The paper tackles learning region embeddings for urban prediction by integrating demographic distributions with mobility and POI signals to capture regional context. It introduces a multi-view framework built around Heterogeneous Region Embedding (HRE), employing a relation-aware GCN, self-attention, and a fusion layer to produce region embeddings $\mathcal{E} = \{\vec{e}_1, \dots, \vec{e}_n\}$ in $\mathbb{R}^d$, trained with a multi-task objective that includes a demographic term $L_{demo}$ alongside $L_n$, $L_{poi}$, and $L_{mobility}$. To balance contributions across views, it adopts Jensen-Shannon divergence $D_{JS}$, defined as $D_{JS}(p\parallel q) = \frac{1}{2} D_{KL}(p \parallel \frac{p+q}{2}) + \frac{1}{2} D_{KL}(q \parallel \frac{p+q}{2})$, which is symmetric and bounded, improving training stability. Empirical results on NYC and CHI show that incorporating income yields up to $10.22\%$ improvements in downstream tasks (check-in, crime, house price) over baselines, with Mobility + Income typically strongest. In mobility-scarce settings, the authors propose Income + Neighbor as a practical alternative, illustrating the framework's applicability to developing cities and its potential for transferable urban prediction.

Abstract

Demographic data, such as income, education level, and employment rate, contain valuable information of urban regions, yet few studies have integrated demographic information to generate region embedding. In this study, we show how the simple and easy-to-access demographic data can improve the quality of state-of-the-art region embedding and provide better predictive performances in urban areas across three common urban tasks, namely check-in prediction, crime rate prediction, and house price prediction. We find that existing pre-train methods based on KL divergence are potentially biased towards mobility information and propose to use Jenson-Shannon divergence as a more appropriate loss function for multi-view representation learning. Experimental results from both New York and Chicago show that mobility + income is the best pre-train data combination, providing up to 10.22\% better predictive performances than existing models. Considering that mobility big data can be hardly accessible in many developing cities, we suggest geographic proximity + income to be a simple but effective data combination for region embedding pre-training.

Figures (1)

• Figure 1: Demographic information encoding and the model structure.