Zipei Fan

CitySpectrum: a non-negative tensor factorization approach

People flow at a citywide level is in a mixed state with several basic patterns (e.g. commuting, working, commercial), and it is therefore difficult to extract useful information from such a mixture of patterns directly. In this paper, we proposed a novel tensor factorization approach to modeling city dynamics in a basic life pattern space (CitySpectral Space). To obtain the CitySpectrum, we utilized Non-negative Tensor Factorization (NTF) to decompose a people flow tensor into basic life pattern tensors, described by three bases i.e. the intensity variation among different regions, the time-of-day and the sample days. We apply our approach to a big mobile phone GPS log dataset (containing 1.6 million users) to model the fluctuation in people flow before and after the Great East Japan Earthquake from a CitySpectral perspective. In addition, our framework is extensible to a variety of auxiliary spatial-temporal data. We parametrize a people flow with a spatial distribution of the Points of Interest (POIs) to quantitatively analyze the relationship between human mobility and POI distribution. Based on the parametric people flow, we propose a spectral approach for a site-selection recommendation and people flow simulation in another similar area using POI distribution.

CityMomentum: an online approach for crowd behavior prediction at a citywide level

Human movements are difficult to predict, especially, when we consider rare behaviors that deviate from normal daily routines. By tracing the behavior of a person over a long period, we can model their daily routines and predict periodical behaviors, whereas rare behaviors, such as participating in the New Years Eve countdown, can hardly be predicted readily and thus they have usually been treated as outliers of the daily routines in most existing studies. However, for scenarios such as emergency management or intelligent traffic regulation, we are more interested in rare behaviors than daily routines. Using human mobility Big Data, the rare behavior of each individual in a social crowd is no longer rare and thus it may be predicted when we analyze the crowd behavior at a citywide level. Therefore in this study, instead of predicting movement based on daily routines, we make short-term predictions based on the recent movement observations. We propose a novel model called CityMomentum as a predicting-by-clustering framework for sampling future movement using a mixture of multiple random Markov chains, each of which is a Naive Movement Predictive model trained with the movements of the subjects that belong to each cluster. We apply our approach to a big mobile phone GPS log dataset and predict the short-term future movements, especially during the Comiket 80 and New Years Eve celebration. We evaluate our prediction by a Earth Mover Distance (EMD) based metric, and show our approach accurately predicts the crowd behavior during the rare crowd events, which makes an early crowd event warning and regulation possible in the emergent situations.

Comparative Perspective of Human Behavior Patterns to Uncover Ownership Bias among Mobile Phone Users

With the rapid spread of mobile devices, call detail records (CDRs) from mobile phones provide more opportunities to incorporate dynamic aspects of human mobility in addressing societal issues. However, it has been increasingly observed that CDR data are not always representative of the population under study because it only includes device users alone. To understand the discrepancy between the population captured by CDRs and the general population, we profile principal populations of CDRs by analyzing routines based on time spent at key locations and compare these data with those of the general population. We employ a topic model to estimate typical routines of mobile phone users using CDRs as topics. The routines are extracted from field survey data and compared between those of the general population and mobile phone users. We found that there are two main population groups of mobile phone users in Dhaka: males engaged in an income-generating activity at a specific location other than home and females performing household tasks and spending most of their time at home. We determine that CDRs tend to omit students, who form a significant component of the Dhaka population.

Deep ROI-Based Modeling for Urban Human Mobility Prediction

Rapidly developing location acquisition technologies have provided us with big GPS trajectory data, which offers a new means of understanding peoples daily behaviors as well as urban dynamics. With such data, predicting human mobility at the city level will be of great significance for transportation scheduling, urban regulation, and emergency management. In particular, most urban human behaviors are related to a small number of important regions, referred to as Regions-of-Interest (ROIs). Therefore, in this study, a deep ROI-based modeling approach is proposed for effectively predicting urban human mobility. Urban ROIs are first discovered from historical trajectory data, and urban human mobility is designated using two types of ROI labels (ISROI and WHICHROI). Then, urban mobility prediction is modeled as a sequence classification problem for each type of label. Finally, a deep-learning architecture built with recurrent neural networks is designed as an effective sequence classifier. Experimental results demonstrate that the superior performance of our proposed approach to the baseline models and several real-world practices show the applicability of our approach to real-world urban computing problems.

Online Deep Ensemble Learning for Predicting Citywide Human Mobility

Predicting citywide human mobility is critical to an effective management and regulation of city governance, especially during a rare event (e.g. large event such as New Years celebration or Comiket). Classical models can effectively predict routine human mobility, but irregular mobility during a rare event (precedented or unprecedented), which is much more difficult to model, has not drawn sufficient attention. Moreover, the complexity and non-linearity of human mobility hinders a simple model from making an accurate prediction. Bearing these facts in mind, we propose a novel online gating neural network framework with two phases. In the offline training phase, we train a gated recurrent unit-based human mobility predictor for each day in our training set, while in the online predicting phase, we construct an online adaptive human mobility predictor as well as a gating neural network that switches among the pre-trained predictors and the online adaptive human predictor. Our approach was evaluated using a real-world GPS-log dataset from Tokyo and Osaka and achieved a higher prediction accuracy than baseline models.