Predicting Economic Growth by Region Embedding: A Multigraph Convolutional Network Approach

With the rapid progress of global urbanization and function division among different geographical regions, it is of urgent need to develop methods that can find regions of desired future function distributions in applications. For example, a company tends to open a new branch in a region where the growth trend of industrial sectors fits its strategic goals, or is similar to that of an existing company location; while a job hunter tends to search regions where his/her expertise aligns with the industrial growth trend providing sufficient job opportunities to sustain future employment and job-hopping.

Our solution is to learn a distribution (aka. embedding) of the growth of various industrial sectors for each region, so that the embeddings of different regions can be searched, or compared for similarity querying. We consider the fine granularity of ZIP code areas as they are usually representative of the regional functions. By effectively utilizing open data on the Internet such as government data (e.g., from US Census Bureau) and third-party data for supervised learning, we propose to first construct a multigraph that captures the various relationships between regions such as direct flight connections and shared school districts, and then learn region embeddings using a novel graph convolutional network architecture. Our multigraph convnet (MGCN) differentiates various feature types such as demographic, social, economic and housing features, and learns different weights on different features and spatial relationships for effective data-driven feature aggregation.

While deep learning is known to require large amounts of data to train, our weighted MGCN (WMGCN) is designed to minimize the number of parameters so that it does not underfit on the limited amount of open data. Extensive experiments are conducted to compare our WMGCN model with several competitive baselines to demonstrate the superiority of our WMGCN design.