Personalized location prediction for group travellers from spatial–temporal trajectories

https://doi.org/10.1016/j.future.2018.01.024Get rights and content

Highlights

  • Investigation of the location prediction problem with consideration of travel group types.

  • Proposing a novel group discovery approach taking the people movement behaviour into consideration.

  • Proposing a classification model to predict the type of the discovered groups.

  • Proposing a novel location prediction model taking the general and group based rules into account.

Abstract

In recent years, research on location predictions by mining trajectories of users has attracted a lot of attentions. Existing studies on this topic mostly focus on individual movements, considering the trajectories as solo movements. However, a user usually does not visit locations just for the personal interest. The preference of a travel group has significant impacts on the places they visit. In this paper, we propose a novel personalized location prediction approach which further takes into account users’ travel group type. To achieve this goal, we propose a new group pattern discovery approach to extract the travel groups from spatial–temporal trajectories of users. Type of the discovered groups, then, are identified through utilizing the profile information of the group members. The core idea underlying our proposal is the discovery of significant movement patterns of users to capture frequent movements by considering the group types. Finally, the problem of location prediction is formulated as an estimation of the probability of a given user visiting a given location based on his/her current movement and his/her group type. To the best of our knowledge, this is the first work on location prediction based on trajectory pattern mining that investigates the influence of travel group type. By means of a comprehensive evaluation using various datasets, we show that our proposed location prediction framework achieves significantly higher performance than previous location prediction methods.

Introduction

With the rapid development of mobile devices and location acquisition technologies, an enormous amount of trajectory data recording the movement of people is available. These overwhelming amounts of data is tremendously useful for the rapidly growing location-based applications market. Due to various requirements of these applications, e.g., system efficiency and marketing efficacy, accurately predicting the next location to which a user may move is essential. The location prediction technique identifies the next location that is most likely to be visited by the user, according to a set of application-dependent locations or pre-determined locations. By knowing the next movement of users, resources can be efficiently allocated to the most possible location, rather than the blind resource allocation. Efficient resource allocation to mobile users would lead to higher resource utilization and lower latency in accessing the resources. In addition, predicting the subsequent location can provide the insights for many existing pervasive applications, such as targeted advertising and services recommendation [1].

The problem of predicting the next location where a user will move has received many research interests in recent years. As the location prediction process is very similar to the location recommendation, many existing works [[2], [3], [4]] intuitively applied a location recommendation approach as their location prediction model. However, there are a few difficulties in adopting the location recommendation for the location prediction. First, the location recommendation process is a non-real-time estimation which means that the recent movements of the user are not taken into account in making the recommendations. Second, conventional location recommendation methods only consider the interest of the user such that they suggest a new location that a user may be interested in. However, the problem of next location prediction focuses on inferring the next location that a user will visit which not only considers the user’s interest, but also the intention of the user. People do not solely visit locations because they are interested, they also go to places because they have to. Consequently, it is not straightforward to apply these recommendation techniques in location prediction.

On the other hand, considering the fact that human movement exhibits sequential patterns, various sequential pattern mining techniques [[5], [6], [7], [8]] have been developed for location predictions. These approaches address the location prediction as a historical movement matching problem. They usually consider the user’s movement trajectory as a sequence of locations, and then, extract the frequent movement patterns from the set of trajectories. These frequent patterns then will be used as the prediction rules to be matched with the previous movement of the user. The difference of these approaches is mainly about the type of the movement pattern they discover. However, they did not take the personalization into account as their approaches just return the same sequence patterns for all the users.

To extract the significant movement patterns, the existing methods mine the frequent sequences of locations from individual user trajectories, such that they assume all the movements as the solo movements. Accordingly, the extracted movement patterns only reflect the individual intention/interest. However, it has been shown that people do not visit locations just for their personal intention/interest. They also go to places which are motivated by the group’s intention/interest they travel with [[9], [10]]. The preferences of the travel group, which may comprise very diverse people, have significant impacts on the places that users visit. Taking a family comprising two adults and a child who walking in a shopping mall as an example, considering only the individuals, i.e. parents, may lead to the different location prediction than when the group type, family, is taking into account.

In this paper, we propose a personalized framework to predict the next location of the users. The core idea underlying our proposal is the discovery of significant movement patterns by considering not only the individual movements, but also the group movements. We first, extract the groups of people who travel together, group travellers, from the spatial–temporal trajectories. Second, the profile information of the users will be used in order to identify the type of the extracted group. Third, the significant movement patterns will be discovered taking into account the group specific movements and individual movements. Finally, the problem of location prediction will be formulated as an estimation of the probability of a given user visiting a given location based on his/her current movement and his/her group type.

In order to extract the group travellers from spatial–temporal trajectories, we propose a novel group pattern, Loose Travelling Companion Pattern (LTCP), with taking into account the properties of human movement behaviour. Extracting the significant movement rules to support the prediction model is also a critical and challenging issue. We define two categories of significant movement rules: General sequential rules (SR) which refers to the movement rules considering the individual movement pattern, and Group-based sequential rules (GSR) which refers to movement patterns associated with the group types. Discovered movement rules then are utilized to construct the prediction model with further incorporating the distribution of places and group types.

The main contributions of this paper are summarized below:

  • To the best of our knowledge, this is the first work that investigates the location prediction problem with consideration of group movement;

  • We propose a novel group discovery approach to identify the groups of people who move together considering the human movement behaviour;

  • We apply a classification model to predict the type of the discovered groups, utilizing the profile information of the users;

  • We propose a novel location prediction model that takes into account the general movement rules and the group-based movement rules to predict next location of the user;

  • We present comprehensive experimental results over various datasets. The results demonstrate that our proposed framework significantly outperforms the widely used sequential prediction technique.

The remainder of this paper is organized as follows. First, we briefly review the related work in Section 2 and present an overview of our proposed prediction framework in Section 3. Next, our proposed group pattern discovery approach and group type prediction technique are described in detail through Sections 4 Group discovery from people movement trajectories, 5 Predicting travel group types, respectively. The movement rules are discovered in Section 6, and the prediction model is constructed in Section 7. A real case study is introduced in Section 8. The performance of our proposal through an empirical evaluation study is discussed in Section 9. Finally, conclusion and directions for the future works are given in Section 10.

Section snippets

Related work

The problem of predicting the future location has been variously formulated in the literature. The first strategy is Vector Based Prediction model which estimates the object’s future location through applying the motion functions. These approaches can be divided into two types: (1) linear models assume that object’s movement follows a linear pattern [[11], [12], [13], [14], [15]], and (2) non-linear models, on the other hand, take into consideration both linearity and non-linear patterns in

System overview

Fig. 1 shows the architecture of our system, which is divided into two main parts, “Offline” and “Online”. In the offline part, the prediction rules are minded from the existing movement trajectories, and through the online part, the location prediction for the incoming trajectory will be done. The groups of people are firstly detected from the movement trajectories (Section 4). The individual information of group members then is used for conducting the travel group type prediction (Section 5).

Group discovery from people movement trajectories

Several studies have been proposed in the literature to discover the groups of moving objects. The main focus of the previous approaches, however, is on the movement trajectories of vehicles or animals with the aim of finding the general trends [[45], [46]]. Different from them, we concentrate on human movement trajectories, particularly in the indoor environment. The movement of people is rather different from the movement of animals or vehicles. People might belong to the same group, while

Predicting travel group types

It has been pointed out that the preferences of the group people travel with have a significant impact on the locations they visit. Taking a family group and a couple group as examples, predicting a location preferred by the family to both groups may not satisfy the couple preferences.

Through the previous step, we discovered the traveller groups from the movement trajectory data. In this step, we identify the type of the discovered groups by utilizing the profile information, Id, of the users.

Group-specific pattern mining

Through the previous sections, traveller groups and their types are discovered. In this section, we use this information to discover the significant movement patterns, the patterns which are frequently visited by the users. We propose a new type of pattern, called GFSP (Group-Based Frequent Sequential Pattern), to represent the frequent movement behaviours by considering the group travellers. In contrast to the conventional sequential pattern, which considers the individual movements, we take

Group-specific location prediction model

According to Fig. 1, the input of the prediction framework include the target user’s profile composed of the group he/she belongs, historical movements (sequence of the visited locations), and the output is the most probable location he/she is going to visit. First, the group that the user belongs to is discovered through applying the group discovery approach, as described in Section 4. Then, the type of the group (family, friends, couple, or solo traveller) will be obtained by applying the

Charlie: Smart trolley

Over the last decade, airport industry has evolved dramatically to a commercial enterprise with the focus on the importance of customers. In large airports, like Heathrow Airport in London, thousands of passengers are served every day. Providing the right services to the passengers is the ultimate goal for the airport authority. In this regard, access to the passenger information plays an important role.

Charlie is a smart trolley developed by working with industrial partner Wuxi Chigoo

Experiments

In this section, we conducted a series of experiments to evaluate the performance of the proposed prediction model, under various conditions. Experiments can be divided into three parts: (1) group discovery evaluation, (2) group type prediction evaluation, and (3) next location prediction evaluation. All of the experiments were implemented in Python on a 3.30 GHz machine with 4 GB of memory running Ubuntu.

Conclusion

In this paper, we defined a new kind of frequent pattern, namely GFSP Pattern, which takes into account the group travel type of the users. Accordingly, we proposed a novel personalized prediction framework to predict the next location of a user for applications such as location-based services. The core idea of our prediction module is a novel prediction strategy that evaluates the score of the next location for a given user by mining the movement patterns of users in terms of the general and

Acknowledgements

The authors would like to thank Wuxi Chigoo Interactive Technology Co. Ltd to sponsor a research project to conduct research in this area, provide data for analysis and also implement and test the algorithms. The authors would also like to thank joint support from Royal Society of Edinburgh under grant number NNS/INT 15-16 Casaseca and National Natural Science Foundation of China under grant number 6151101271 to engage academic staff from the UK and China to work together on this project. In

Elahe Naserian is currently working towards the Ph.D. degree from the School of Engineering and Computing, University of the West of Scotland, United Kingdom. She obtained her master’s degree from the University of Tehran, Iran. Her research interests include data mining and knowledge discovery over spatial–temporal data.

References (49)

  • YavasG. et al.

    A data mining approach for location prediction in mobile environments

    Data Knowl. Eng.

    (2005)
  • TsengV.S. et al.

    Efficient mining and prediction of user behaviour patterns in mobile web systems

    Inf. Softw. Technol.

    (2006)
  • PengW.C. et al.

    Developing data allocation schemes by incremental mining of user moving patterns in a mobile computing system

    IEEE Trans. Knowl. Data Eng.

    (2003)
  • Y. Ge, H. Xiong, A. Tuzhilin, K. Xiao, M. Gruteser, M.J. Pazzani, An energy-efficient mobile recommender system, in:...
  • Q. Liu, Y. Ge, Z. Li, E. Chen, H. Xiong, Personalized travel package recommendation, in: proceeding of ICDM,...
  • J. Zhuang, T. Mei, S. Hoi, Y. Xu, S. Li, When recommendation meets mobile: Contextual and personalized recommendation...
  • MorzyM.

    Prediction of moving object location based on frequent trajectories

  • MorzyM.

    Mining frequent trajectories of moving objects for location prediction

  • YingJ.J. et al.

    Mining geographic-temporal-semantic patterns in trajectories for location prediction

    ACM Trans. Intell. Syst. Technol.

    (2013)
  • A.J. Cheng, Y.Y. Chen, Y.T. Huang, W.H. Hsu, H.Y.M. Liao, Personalized travel recommendation by mining people...
  • ChenY.Y. et al.

    Travel recommendation by mining people attributes and travel group types from community-contributed photos

    IEEE Trans. Multimedia

    (2013)
  • J.M. Patel, Y. Chen, V.P. Chakka, Stripes: an efficient index for predicted trajectories, in: SIGMOD, 2004, pp....
  • C.S. Jensen, D. Lin, B.C. Ooi, Query and update efficient B+-tree based indexing of moving objects, in: VLDB, 2004, pp....
  • S. Saltenis, C.S. Jensen, S.T. Leutenegger, M.A. Lopez, Indexing the positions of continuously moving objects, in:...
  • Y. Tao, D. Papadias, J. Sun, The tpr*-tree: An optimized spatio-temporal access method for predictive queries, in:...
  • S.M. Ghoreyshi, A. Shahrabi, T. Boutaleb, An underwater routing protocol with void detection and bypassing capability,...
  • Y. Tao, C. Faloutsos, D. Papadias, B. Liu, Prediction and indexing of moving objects with unknown motion patterns, in:...
  • C.C. Aggarwal, D. Agrawal, On nearest neighbor indexing of nonlinear trajectories, in: PODS, 2003, pp....
  • SongC. et al.

    Limits of predictability in human mobility

    Science

    (2010)
  • A. Asahara, A. Sato, K. Maruyama, K. Seto, Pedestrian-movement Prediction based on Mixed Markov-chain model, in: Proc....
  • AshbrookD. et al.

    Using gps to learn significant locations and predict movement across multiple users

    Pers. Ubiquitous Comput.

    (2003)
  • A. Bhattacharya, S.K. Das, LeZi-update: An information-theoretic approach to track mobile users in PCS networks, in:...
  • Y. Ishikawa, Y. Tsukamoto, H. Kitagawa, Extracting mobility statistics from indexed spatio-temporal dataset, in: Proc....
  • F. Giannotti, M. Nanni, F. Pinelli, D. Pedreschi, Trajectory pattern mining, in: Proceedings of the 13th ACM SIGKDD...
  • Cited by (26)

    • A survey on next location prediction techniques, applications, and challenges

      2022, Eurasip Journal on Wireless Communications and Networking
    View all citing articles on Scopus

    Elahe Naserian is currently working towards the Ph.D. degree from the School of Engineering and Computing, University of the West of Scotland, United Kingdom. She obtained her master’s degree from the University of Tehran, Iran. Her research interests include data mining and knowledge discovery over spatial–temporal data.

    Xinheng Wang (M’04-SM’14) received the B.E. and M.Sc. degrees in electrical engineering from Xian Jiaotong University, Xian, China, in 1991 and 1994, respectively, and the Ph.D. degree in electronics and computer engineering from Brunel University London, Uxbridge, U.K., in 2001. He is currently a Professor of Computing with the School of Computing and Engineering, University of West London, London, U.K. He holds 6 granted and further 12 published invention patents and has authored or co-authored over 70 referred journal papers. He has broad research experience in mobile healthcare, asset monitoring, and wireless mesh and sensor networks, where the technologies developed in wireless mesh networks have been commercialized at Swanmesh Networks Ltd (www.swanmesh.com). His current research interests include indoor positioning, Internet-of-Things, and big data analytics for smart airport services, where he has developed the world’s first smart trolley with industry partner.

    Keshav Dahal is a Professor of Intelligent Systems and the leader of the Artificial Intelligence, Visual Communication and Network (AVCN) Research Centre at the University of the West of Scotland (UWS), UK. He is also affiliated with Nanjing University of Information Science and Technology (NUIST) China. Before joining UWS he was with Bradford and Strathclyde Universities in UK. He obtained his Ph.D. and Master degrees from the University of Strathclyde, UK. His research interests lie in the areas of applied AI to intelligent systems, trust and security modelling in distributed systems, and scheduling/optimization problems. He has published extensively with award winning papers, and has sat on organizing/program committees of over 60 international conferences including as the General Chair and Programme Chair. He is a senior member of the IEEE.

    Zhi Wang is currently an associate professor at State Key Laboratory of Industrial Control, Zhejiang University, China. His main research focus is on acoustic signal and array processing, sparsity signal and compressive sensing, localization and tracking of mobile target, multiple sensor fusion, crowd-sensing and mobile computing, big-data and industrial IoT protocols. He has co-authored over 100 publications in international journals and conferences and also has served as an member of advisory board and an editor of several conferences. He is also the committee member for China Computer Federation Sensor Network Technical Committee and China National Technical Committee of Sensor Network Standardization, and is a member of IEEE and ACM.

    Zaijian Wang received his B.E. degree (2002) from Anhui Polytechnic University, M.Sc. degree (2005) from University of Science and Technology of China, and Ph.D. degree (2015) from Nanjing University of Posts and Telecommunications. His current research interests focus on multimedia big data, end-to-end QoS provisioning and wired/wireless multimedia streaming. He is currently an associate professor at College of Physics and Electronic Information, Anhui Normal University, Wuhu, China.

    View full text