MIAC: A mobility intention auto‐completion model for location prediction

• DOI: http://doi.org/10.1002/isaf.1432
• Author: Zhi Li,Limin Sun,Guan Feng,Feng Yi,Hongtao Wang
• Published date: 01 October 2018
• Date: 01 October 2018

Received: 13 March 2018 Revised: 19 May 2018 Accepted: 22 May 2018

DOI: 10.1002/isaf.1432

RESEARCH ARTICLE

MIAC: A mobility intention auto-completion model for location

prediction

Fen g Yi 1,2 Guan Feng3Hongtao Wang4Zhi Li2Limin Sun2

1School of Computer Engineering, University

of Electronic Science and Technology of China,

Zhongshan Institute No. 1 xueyuanRoad, ShiQi

District, Zhongshan, China

2Institute of Information Engineering, Chinese

Academyof Sciences, Beijing 100093, China

3Economics and Business administration,

ChongQing University,No.174 Shazhengjie,

Shapingba, Chongqing, 400044, China

4School of Control and Computer Engineering,

North China Electric PowerUniversity, Baoding

071051, China

Correspondence

Guan Feng,Economics and Business

administration,ChongQing University, No.174

Shazhengjie, Shapingba, Chongqing, 400044,

China.

Email: 516234682@qq.com

Fundinginformation

National NaturalScience Foundation of China,

Grant/AwardNumber: 61472418; Major R&D

Plan, Grant/AwardNumber: 2016YFC1202204

Handling Editor: O'Leary Daniel

Summary

Location prediction is essential to many commercial applications and enables appealing expe-

rience for business and governments. Many research work show that human mobility is highly

predictable. However, existing work on location prediction reported limited improvements in

using generalized spatio-temporal features and unsatisfactory prediction accuracyfor complex

human mobility. To address these challenges, in this paper we propose a Mobility Intention and

Auto-Completion (MIAC) model. We extract those mobility patterns that generalize common

spatio-temporal features of all users, and use the mobility intentions as the hidden states from

mobility dataset. A new predicting algorithm based on auto-completion is then proposed. The

experimental results on real-world datasets demonstrate that the proposed MIAC model can

properly capture the regularity of a user's mobility by simultaneously considering the spatial and

temporal features. The comparison results also indicate that MIAC model significantly outper-

forms state-of-the-art location prediction methods, and also can predicts long rangelocations.

KEYWORDS

auto-completion, data mining, location prediction, spatiotemporaldata

1INTRODUCTION

Along with the popularity of smart devices with sensing technology

(Caiet al., 2016), the past decade has witnessed a tremendous increase

in the availability of mobility data, ranging from early cellular tower

data of Personal Communication Systems (PCS) (Yavas et al., 2005), GPS

trajectories (Mathew et al., 2012) to the check-in data of various

location-based services (LBSs) (Hasan et al., 2013). In addition, systems

originally designed for fare purpose are also enriched with mobility

data, such as the Smart Card Data (SCD)for public transportation (Itoh

et al., 2014). Those available massive mobility data have profoundly

changed the waypeople think, behave and interact, in activities such as

trip scheduling, activity planning and friends making. They can be uti-

lized to model human mobility for the spread prediction of epidemic

diseases (Colizzaet al., 2007) or for the communities detection (McGee

et al., 2013).

Recently, location prediction is becoming an important research

topic for human mobility, and it is considered as the core function

in various commercial applications, such as mobile marketing (Barnes

& Scornavacca, 2004), tourism market (Napoli et al., 2017), advertis-

ing and recommendation etc.. Location prediction can support vari-

ous context-awareapplications. For example,when knowing where the

crowd will gather at a given time, merchants can prepare appropriate

merchandise or service in advance. Especially,in important festivals, if

the crowded locations can be predicted accurately, the relevant busi-

ness services can be directed towards those locations. In case of trans-

portation, accommodation, shopping or catering, goods and services

can be more efficiently allocated to the most needed locations.

For its potential value, location prediction has attracted significant

research efforts in past few years. Almost all location prediction mod-

els are based on the mobility regularity embedded in history mobility

data, and there are two essential sub tasks involved:howtodiscoverand

represent the mobility regularity;andhowto utilize the mobility regularity

for prediction. Traditional location prediction models usually represent

themobility regularity as spatiotemporal features (Ashbrook & Starner,

2003; Song et al., 2004). By assuming the human mobility patterns as

uniform in consecutive number of appearances, most prediction algo-

rithms can then be based on Markovmodels.

Intell Sys Acc Fin Mgmt. 2018;25 161–173. wileyonlinelibrary.com/journal/isaf ©2018 John Wiley & Sons, Ltd. 161: