Demo2Vec: Learning Region Embedding with Demographic Information

Abstract

<p>Demographic data, such as income, education level, and employment rate, contain valuable information about urban regions, yet few studies have integrated demographic information to generate<br>region embedding. In this study, we show how simple and easy-to-access demographic data can enhance the quality of state-of-the-art region embedding and improve predictive performance in urban<br>areas across three common urban tasks, namely check-in prediction, crime rate prediction, and house price prediction. We find that existing pre-training methods based on KL divergence are potentially<br>biased towards mobility information and propose using Jenson-Shannon divergence as a more appropriate loss function for multi-view representation learning. Experimental results from both New York and Chicago indicate that Mobility + Income is the best pre-training data combination, providing up to 10.22% improvement in predictive performance compared to existing models. Since mobility big data is hard to access in many developing cities, Income + Neighbor may serve as an alternative data combination for<br>effective region embedding learning. <br></p>