Revealing functional regions via joint matrix factorization based model

Various functions in the region are emerging with the process of urbanizations, such as the residential, entertainment or hospital districts, which can be indicated during the urban planning. Recently, researchers try to discover the region functions with the human mobility data based on machine learning and statistical models. However, previous work always employs the single domain data like mobility information or district attributes to measure the functions. In this paper, we will address the problem by integrating multi-domain data, like human trajectories, base station information and Points-of-Interest attributes. For that, we propose a joint nonnegative matrix factorization model to combine the multi-source data and extract the function distribution of each urban region, then the dominated function among the regions can be uncovered based on the clustering process over the extracted distributions. We also evaluate the performance of our method on the real-world dataset to demonstrate the advantages of our proposed model over the baseline methods.