A Multi-layer Data Representation of Trajectories in Social Networks based on Points of Interest

Transcription

1 A Multi-layer Data Representation of Trajectories in Social Networks based on Points of Interest Reinaldo Bezerra Braga LIG UMR 5217, UJF-Grenoble 1, Grenoble-INP, UPMF-Grenoble 2, CNRS 38400, Grenoble, France Michela Bertolotto School of Computer Science and Informatics, University College Dublin (UCD) Dublin, Ireland Ali Tahir School of Computer Science and Informatics, University College Dublin (UCD) Dublin, Ireland Hervé Martin LIG UMR 5217, UJF-Grenoble 1, Grenoble-INP, UPMF-Grenoble 2, CNRS 38400, Grenoble, France ABSTRACT Social networking, and sophisticated wireless and positioning systems are fast developing and ever increasing technologies. Mobile social applications have the ability to increase the social connectivity by capturing automatically users daily routines with Global Positioning System (GPS) receivers. These applications allow to record users trajectories based on daily travel routes as well as to share experiences and interests among friends. However, there is always an increasing demand for providing an easy way to manipulate trajectory data, to generate and compare user profiles. Effective analysis of spatial trajectories has become an essential requirement to explore and understand the behavior of moving objects. In this paper, we highlight the importance of capturing users daily routines in the form of trajectories in order to strengthen social connectivity. We also present the conceptual approach to multi-layer data representation in order to extract points of interest of correlated trajectories. Finally, we show how the data model could provide mobile social applications with direct support for trajectories at different abstraction levels. Categories and Subject Descriptors C.0 [Computer Systems Organization]: [General - Modeling of computer architecture]; H.1.2 Information Systems [Models and Principles - User/Machine Systems - Human information processing] General Terms Design, Verification Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. WIDM 12, November 2, 2012, Maui, Hawaii, USA. Copyright 2012 ACM /12/11...$ Keywords Mobile Social Application; Multi-layer Representation; Geographic Information System 1. INTRODUCTION A spatial trajectory can be defined as a trace generated by a moving object in geographical space, which is represented by a sequence of geospatial coordinates and a timestamp [17]. Sharing of trajectory data has increased over the last years due to the availability of sophisticated Web and mobile applications (including social networks). For instance, users can easily capture and share their trajectories over time based on their daily trips. In our previous work [2], we presented a trajectory correlation algorithm based on users daily routines, where the key idea was to recommend content to single users or alternatively to group users together based on similar trajectories and/or preferences. We observed that analyzing user trajectories over time could reveal habits and preferences. This could be exploited in a large number of applications from logistics and transportation, to car-pooling and recommendation systems etc. While the capability to capture a sequence of positions is the starting point of managing movement, designing an application based on trajectory data requires a structural approach. Indeed many applications need a more structured recording of movement and semantics, e.g. as a temporal sequence of journeys, each one occupying a time interval in the object s lifespan and taking the object from a departure point to a destination point [10]. Moving objects can represent vehicles delivering posts within a given region, migration of animals, and a person that goes from home to work and back everyday. Hence, trajectories can be defined in terms of moving objects traveling from a departure point to a destination point during these countable journeys. After capturing these trajectories, modeling them becomes necessary for important operations, such as: i) to indentify patterns, which will be used for decision making (e.g. registering users trajectories within a city for optimizing traffic of vehicles); ii) to query information about the moving objects (e.g. enriching trajectory data with context information); 19

2 iii) to optimize intelligent transport systems (e.g. motivating users to use car pooling alternatives in order to reduce the number of vehicles in urban regions). The main motivations to design a suitable data model are related to providing an easy way to manipulate trajectory data, to use structured query languages, to specify profiles through movements, to create and compare profile groups. On the other hand, the identification of the scenario is a significant requirement to design a conceptual data model. In this paper, we take into account the scenario of an employee that goes from home to work and back everyday within a city, whose trajectory can be represented at different abstraction levels according to a set of Points of Interest (PoI) [12]. In addition, we consider a diversity of semantic data that enriches the knowledge on these trajectories. For a user daily trip, we can obtain information about possible PoI s based on user movements. For example, whenever the user goes from work to home, he usually stops at a specific coffee shop. The conceptual model for trajectories must be able to analyze and manage simple trajectories (direct travels from origin to destination) as well as complex trajectories (where the trajectory is semantically composed of separate segments and/or different abstraction levels). Furthermore, the data model must relate any type of semantic annotation to trajectories, such as attributes of each trajectory and connections between the trajectory and an object stored in the database. Often, it is important to understand the movement data at multiple abstraction levels for pattern recognition and analyzing movement behaviors as well as to deduce the relationships between users in social networks. In order to create a flexible data model for mobile social application context, we propose a multi-layer data representation of moving objects based on PoI s. This model could be employed in the context of mobile social applications. The remainder of this article is organized as follows. To provide the necessary context for our work, we discuss the related work in section 2. A conceptual overview of PoI and multi-layer data representation is presented in Section 3. The multi-layer data representation for our approach is presented in Section 4. Section 5 details the implementation of our multi-layer data representation in a real scenario. Finally, Section 6 presents the conclusions and some directions for future work. 2. RELATED WORK Several researchers have shown an interest in analyzing and representing spatio-temporal data [2][14][16]. This data is relevant in a number of areas such as social interaction, data mining, medicine and geographical information system. For instance, in the context of social interaction, we point out some approaches related to collecting and analyzing daily trajectories of humans, addressing issues such as daily routine, mobility, sport, trips, and social networks. In all these approaches, the amount of data produced is very large and is therefore challenging to interpret. In parallel, the need for representing information about PoI on the Web has emerged [12] in order to manage and organize context-aware information. Interesting issues include how points or regions can be correlated through multi-layer representation [7] and how user trajectories could be analyzed in terms of their distance to another one [13]. Multi-layer data representation has been of interest for a long time due to its importance for spatial data representation [8][6][4]. In spite of the large number of issues about multi-layer data representation, there is a lack of multi-layer representation techniques for moving object trajectories. In [18], the authors present a design for multi-layer spatial objects in which both spatial objects and the vertices of their component geometry are labeled with level priority values. Although the data model supporting queries at different abstraction levels is very interesting, it is not intended for representing trajectories and not easily extendable for this context. In [11], the authors present an interesting Rule-based Location Prediction method (RLP), to guess the user s future location for location-based services. However, they do not consider the partial containment relationship between spatial regions at different spatial levels. In [15] and [5], the authors introduce approaches to consider trajectory patterns between different spatial levels as well as the relation among user, location and trajectory. In particular, GeoLife [15] is a social networking service which increases social connectivity among users taking multiple geospatial scales into account while the work described by [5] focuses on Regions of Interest (ROI) as opposed to multiple abstraction levels. Following these works, we have focused our research on understanding multiple abstraction levels with respect to PoI s in moving object trajectories. We extend our previous work [2][3] and present multi-layer data representation of correlated trajectories (the main contribution of this paper), taking into account the PoI at multiple abstraction levels. This contribution can assist in solving the inherent complexity of trajectories especially at multiple map scales such as trajectory semantics and its relationship with PoI s. 3. CONCEPTUAL VIEW ON POI AND MULTI-LAYER DATA REPRESENTATION With the growth of GPS-embedded modules for mobile devices, large amounts of mobility data are being collected in the form of trajectories. A trajectory data is usually presented as a segment of sample points (TId, lat, lon), where (TiD) is a trajectory identifier and (lat,lon) is a position in space [1]. However, these sample points are usually available with very simple or no semantics. For instance, a trajectory can be a PoI or can have a set of PoI s. In general, a PoI is a location about which information is available. PoI can be represented by an identifier containing a set of coordinates or a three-dimensional model of a building with names in different languages, information about opening hours, and the address. The information of PoI is usually applied in a large number of solutions, such as mapping, navigation systems, location based social networks, networking games, and augmented reality. The W3C Points of Interest Working Group (W3C PoI WG) has defined a specification for PoI data that can be used in a large number of applications [12]. This specification aims at the creation of a flexible, lightweight, extensible PoI data model, as well as a normative syntax for this data model in order to provide best practices for sharing, organizing and serving PoI on the Web. While the PoI data model is an important starting point for the data representation, the relation between data at various levels of abstraction is still a challenge. For instance, the data model has to be able to answer a query with ex- 20

3 User 1 User 2 Trajectories of User 1 Trajectories of User 2 Daily Rou4ne of User 1 Density- Based Clustering Daily Rou4ne of User 2 PoI root PoI root High Layer 1 Abstraction Layer 1 PoI root PoI s, (n+1) PoI s, (n+1) PoI s, n PoI s, n PoI s, (n+1) PoI s, n Layer 2 Layer 2 Low Abstraction PoI PoI (s+1), (n+1) PoI (s+1), n (s+1), n PoI (s+1), (n+1) Layer 3 Layer 3 GPS coordinate Point of Interest (PoI) PoI (s+1), n GPS coordinate PoI (s+1), (n+1) Multi-layer representation of PoI (User 1) Abstraction level of PoI Multi-layer representation of PoI (User 2) Figure 1: PoI in a multi-layer data representation. act knowledge of the data abstraction level, as well as to compute representations of different types of data, taking into account each abstraction level. Therefore, in this paper we consider the impact of multi-layer data on the PoI data representation. The key idea is to extend the relationship element, which is derived from the P OIBaseT ype, taking into account the data model proposed by W3C PoI working group. In our work, a PoI in a multi-layer representation can be represented as shown in Figure 1, where s and n represent the abstraction level and PoI identifier respectively, and the links are the relations between PoI s at different abstraction levels. An example to illustrate a multi-layer data representation can be described through administrative partitions of a territory, such as department, city and local place. In other words, spatial information is described according to some attribute based on the user preference. Following this example, we clearly see that the place is inside the city and the city belongs to the department. In addition, all these locations can be classified as PoI at different abstraction levels. Therefore, we define this spatial information to be a multilayer data representation of each PoI in order to support the trajectory description. Figure 1 presents an example of multi-layer data representation in terms of PoI. According to Figure 1, a PoI is presented by considering its layer. For instance, a PoI for the Layer1 could be represented by a user trajectory (e.g. to go from home to work). The Layer2 is defined as segments that compose the user trajectory, where each segment can be identified by a street name. Finally, the last layer (Layer3) can be represented by the name of the location according to the GPS coordinate (e.g. bakery s name, house number, etc.). Based on this structure, we observe that PoI is presented as a location according to the abstraction level of the user/system. However, a PoI in the high abstraction level may relate to a set of other PoI at low levels. To better understand this relation, we use a tree structure to show the relation between each information according to the multi-layer data representation. According to the middle box shown at the bottom of Figure 1, the tree structure is defined on three levels, where root denotes the trajectory, road segments correspond to the second level, and local places to the third level. We observe that a PoI can be any information presented in this tree, according to a given abstraction level. For example, a user would like to know which of the road segments she takes to go from home to work are similar to those of another friend, who also goes from home to work. In this example, the PoI s of both users are at the same abstraction level (Layer 2) since two road segments are compared. However, the two PoI s could be available at different abstraction levels if a user compares a specific place (Layer 3) with a road segment (Layer 2). With this in mind, the data model has to answer all requests, taking into account each abstraction level. 4. MULTI-LAYER DATA REPRESENTATION BASED ON POI In the previous section we presented the conceptual view on multi-layer data representation based on PoI, in this section we focus on a specific scenario. 21

4 Figure 2: Architecture overview of our approach. 4.1 Scenario In our previous work, we have proposed a novel solution in order to increase social interactions by relating daily routines and PoI based on trajectories of mobile users at a fixed abstraction level [2, 3]. For instance, a mobile social application jointly with a social network can answer the following questions: Which of my friends stop in my preferred coffee shop at the same time of the day? Do any of my friends pass near my apartment to go from their home to their work? Which of my contacts will be passing close to me during the week? In this paper, we extend this work to take different abstraction levels into account. In the following we describe a scenario for our multi-layer data model. The scenario considers users that go from home to work and back everyday. While Figure 1 illustrates the process to register the trajectories and to find a user s daily routine, Figure 2 presents our architecture overview. Mobile users use a mobile social application to register their trajectories in order to describe their daily routines. After visualizing and validating the trajectory that represents users daily routines, the user profile is created and the trajectory information is sent to the structuring module. The structuring module verifies if there is a previous trajectory for the user stored in the database. If there is no trajectory, it creates a new user s daily routine. On the other hand, if multiple trajectories are found, clustering and aggregation techniques are used to identify the aggregated trajectory (a best representative of user s daily routine) [3]. We apply Ordering Points To Identify the Clustering Structure (OP- TICS), a density-based clustering algorithm to classify user trajectories based on their daily routes. The clustering and aggregation module provides the best representative trajectory for each user. This aggregated trajectory from one user is compared to other users by applying our trajectory correlation algorithm [2]. This approach enables groups of users to share similar routes to increase geospatial social interaction. The user daily routine then is enriched with additional information about PoI. The structuring module exports the enriched information to update the user profile database. Based on the illustrated data representation, we design our multi-layer data model for trajectories, taking into account different abstraction levels of PoI. In the following we provide the basic definitions to support our discussion. 1. Trajectory (T ) is defined as a set of consecutive points captured through a GPS of one trip performed by a user. Each location (L) is composed of a set of information (latitude, longitude, altitude, direction, time stamp for each registered point (t L) and an approximate speed provided by the GPS). T = {L 1, L 2, L 3,..., L n}, the time interval between two points is computed by the subtraction of t L(k+1) t L(k), where (1 k < n). Although the points are characterized by latitude, longitude and altitude, we focus on points in 2D space (latitude and longitude) to represent the position of each user. 2. PoI (p) is defined as a human construct [12]. PoI typically denotes a location (L) where a user can identify an entire trajectory, segment of route (e.g. street name) or place, according to the layers presented in Figure 1, typically represented by name and characterized by type, which may be used as a reference point or a target in a location based service request (e.g., route destination). 3. Set of PoI (M) is defined as the set of points of interest (p) based on the abstraction level of multi-layer representation. The set of abstraction levels is represented by (SAL) and each abstraction level (s) is defined as the identifier of a PoI (p) such that s SAL. M is a finite set and subset of other PoI e.g. M = {p 1{p (2,1), p (2,2), p (2,n) }, p 2{p (3,1), p (3,2), p (3,n) },..., p (s 1) {p (s,1), p (s,2), p (s,n) }}, where s is the abstraction level. For instance, the set to represent a user trajectory in the campus of Joseph Fourier University is 22

5 p root SAL 1 p (2,1) p (2,2) p (2,3) SAL 2 p (3,1) p (3,2) p (3,3) p (3,4) p (3,5) p (3,6) p (3,7) SAL 3 Figure 3: Example of multi-layer data representation. M T raj = { Chemistry Street{Grenoble Informatics Laboratory, CERMAV Laboratory}, P iscine Street{ENSIMAG Laboratory}, Library Street{Central Library, Mathe matics Laboratory}} where Traj is the user trajectory in the Layer 1, Chemistry Street, Piscine Street and Piscine Street are road segments in the Layer 2, and Grenoble Informatics Laboratory, CERMAV Laboratory, ENSIMAG Laboratory, Central Library and Mathematics Laboratory are local places in the Layer 3. The intention to design a conceptual model is to offer basic procedures in order to support designers in the development of mobile social applications. A usual feature in the spatial multi-layer data model is the PoI of a user corresponding to a given abstraction level (trajectory, road segment and local place). 4.2 Data representation and abstraction level Taking into account our proposed scenario, we present our data representation based on the three abstraction levels. The set of PoI (M) is then structured by an inclusion relationship, where the partial order is the relation over M. The pair (SAL, ) is used to represent the relation between data, where SAL is a finite and non-empty set of PoI (M) and is a partial order of SAL. Therefore, we note that s (n+1) s n p(s (n+1) ) p(s n), where s represents the abstraction level and p symbolizes the points of interest in that abstraction level. In other words, an abstraction level s (n+1) is a subset of s n, if and only if a point of interest belonging to the abstraction level (p(s (n+1) )) is a partial order of another point of interest at the abstraction level (p(s (n) )). Furthermore, when we consider that a graph of is a tree, we can say that a location L is associated to p SAL in order to indicate that L belongs to the abstraction level s associated to p. Since the multi-layer data representation is presented, we take into account the organization of objects for a defined abstraction level. Consequently, a low abstraction level offers the set of PoI s to describe a user trajectory at the highest abstraction level. We can observe that for all L shown in the data representation, we may have a specific PoI available at each abstraction level (s), such that L p. This representation offers a procedure to understand the set of abstraction levels. Our representation can be applied in the scenario as follows. We consider three main situations of social interaction between users, as shown in Figure 4. For the first situation (Figure 4(a)), we observe that two users have a point of interaction at the crossing of two PoI s (e.g. road segment). In Figure 4(b) the point of interaction could be the complete set of PoI (e.g. all road segment or a part of it). Finally, (Figure 4(c)) there is no interaction between users through PoI s. However, the users can consider a possible social interaction due to the proximity of their trajectories. 4.3 Multi-layer representation of correlated trajectories As each abstraction level contains the set of PoI s to describe a user trajectory, we need to consider each set of abstraction level (SAL). We define a user trajectory as a sequence of PoI s, where the set of trajectories crosses between different abstraction levels in the required order. The following example presents a multi-layer representation in order to illustrate our approach. SAL 1 = {p (root) } SAL 2 = {p (2,1), p (2,2), p (2,3) } SAL 3 = {p (3,1), p (3,2), p (3,3), p (3,4), p (3,5), p (3,6), p (3,7) } Then, we can construct the following sets of PoI (M): M 1 = {p (root) {p (2,1) {p (3,1), p (3,2) }}} M 2 = {p (root) {p (2,2) {p (3,3), p (3,4), p (3,5) }}} 23

6 User A User B User A User B (a) Crossing. (b) Sharing a PoI. User A User B (c) Nearing. Figure 4: Three main situations to consider an approximation between two users. M 3 = {p (root) {p (2,3) {p (3,6), p (3,7) }}} The Figure 3 illustrates these sets of PoI s related to each SAL. In the next definition, the user trajectory descriptor (D) contains the sequence of PoI registered by the user. For instance, we determine two different trajectory descriptors: D 1 = < p (2,1), p (2,2), p (3,7) > D 2 = < p (3,1), p (3,3), p (3,5) > We observe that the descriptors can be composed by PoI s at different abstraction levels due to multiple location names obtained from reverse geocoding services. Therefore, our data representation is able to find a similarity although the PoI s are at different abstraction levels. The concept of multi-layer representation is an important step to understand the relations and similarities between PoI s grouped in different trajectory descriptors. For instance, if we consider D 1, the user describes a trajectory from a departure PoI (at the second abstraction level) to a destination PoI (at the third abstraction level). In case of D 2, the user describes his/her trajectory at the same level. A multi-layer data representation should be able to identify the abstraction level of each PoI in order to find the similarities between different user trajectories. If we observe the previous trajectory descriptors and the three situations presented in Figure 4, we detect some challenges to develop a data model at different abstraction levels. For instance, if we observe D 1 and D 2, we see that the first user is passing in p (2,2) (at the second level) and the other user is passing in p (3,3) (at the third abstraction level). Therefore, we are able to find similar locations between users who are sharing PoI s. 5. CASE STUDY The current multi-layer data representation has been implemented as a prototype for preliminary evaluation. This paper provides a generic data model that can be used in different kinds of applications. As illustrated in Figure 3, we follow a top down approach in order to find similar PoI s between users. Firstly, we compare the highest abstraction levels of both users, taking into account the region around each trajectory. Since we find Figure 5: model. EER Diagram of our multi-layer data the correlated regions of both trajectories, we perform the comparison in the next layer for finding similar road segments between users trajectories, which allows to obtain more details about the type of similarity (e.g. near, sharing). Finally, we carry out the comparison at the lowest abstraction level in order to find similarities between local places, such as: bakery X, hospital Y, and others. Therefore, due to this top down processing, our multi-layer data representation can guarantee the occurrence of any of the three situations presented in Figure 4. Our approach was implemented by using the multi-layer data representation taking into account optimization techniques to find similarities between user trajectories in social networks. An Extended Entity-Relationship (EER) model is shown in Figure 5. The model was generated through MySQL Workbench, a native ER diagramming tool of MySQL [9]. The lines between boxes are arranged according to the Crow s database notation. Basically, the end of each line is shaped to symbolize the relationship cardinality. We generated an optimized data model to find similar PoI s between users based on their trajectories. Figure 5 is useful to illustrate the relations between PoI s at multiple abstraction levels. According to the proposed data model, a user profile in a social network is related to ShareFriend (friends that share their trajectories) and UsualFriend (friends that are able to share their trajectories). In addition, the user profile is linked to Collection, which identifies the trajectory registered by the mobile device, and Correlation, which provides all similar PoI s that were identified by trajectory correlation algorithm. The multi-layer approach is directly related to PoI and TrackPoint tables. When a location or PoI is present in a query, the database is able to identify the set of PoI and abstraction level. According to our algorithm [2], the location or PoI is present in the database, the system will return the accurate information about the situations presented in Figure 4. Figure 6 shows the trajectory of user A in comparison to 24

7 Sharing a PoI Sharing a PoI Near points PoIs out of contact area Figure 7: Best representative PoI (Grenoble) of user B in comparison to user A at a different abstraction level. Figure 6: Best representative trajectory (in Grenoble) of user A in comparison to another user at a fixed abstraction level [2]. a multi-layer data model to support the knowledge of trajectory semantics. This contribution can open interesting research directions such as the analysis of requirements for mobile social applications according to their data representation, which we aim to address in the near future. In particular we intend to investigate other data mining techniques in order to observe their impact in our proposal. We also plan to investigate techniques for indexing of trajectories for efficiency such as text-based mining. Additionally, we aim to reuse our techniques in different types of scenarios (for example car pooling and tourism related applications). Finally, we intend to propose a general framework for the development of context-aware systems based on trajectory correlation, focusing on the impact of sharing trajectory information in online social networks as well as their privacy implications. another user at a fixed abstraction level. We used the colors to represent similar PoI s between them. The green color represents the same PoI (e.g. street name) that is used by both users for their daily routines. The blue color denotes the near PoI s. Finally, the red color indicates the points that are out of the correlated area. Figure 7 illustrates the same comparison, but at a different (less detailed) abstraction level. User B indicates Grenoble as a PoI to find the similarities in comparison to the PoI of user A. As the PoI of user A is a subset of the set of PoI Grenoble, the map is shown with a green dot over Grenoble. Figure 7 presents an example of how a multi-layer data model can provide information at different abstraction levels. Our algorithm allows to identify similar user interests in different abstraction levels due to the top down processing. Firstly, we compare the highest abstraction levels of both users, taking into account the region around each trajectory. Since we find the correlated regions of both trajectories, we perform the comparison in the next layer for finding similar road segments between users trajectories, which allows to obtain more details about the type of similarity (e.g. near, sharing). Finally, we carry out the comparison at the lowest abstraction level in order to find similarities between local places, such as: bakery X, hospital Y, and others ACKNOWLEDGMENTS Research presented in this paper was funded by a Strategic Research Cluster grant (07/SRC/I1168) by Science Foundation Ireland (SFI) under the National Development Plan, the IRCSET Ulysses program, French Ministry of Higher Education and Research, E GIDE program and European Cooperation in Science and Technology (COST). The authors gratefully acknowledge this support. 8. REFERENCES [1] V. Bogorny and M. Wachowicz. A framework for context-aware trajectory. Data Mining for Business Applications, pages , [2] R. B. Braga, S. d. M. Medeiros da Costa, and H. Martin. A trajectory correlation algorithm based on users daily routines. In Proceedings of the 19th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, GIS 11, pages , New York, NY, USA, ACM. [3] R. B. Braga, A. Tahir, M. Bertolotto, and H. Martin. Clustering user trajectories to find patterns for social CONCLUSION AND FUTURE WORK In this paper we present a flexible multi-layer data model for mobile social application context based on PoI. We design a conceptual view to be adaptable and acceptable to a set of generic features as well as to assist developers in designing solutions with the inherent complexity of trajectory semantics. Besides that, we discuss how our trajectory modeling could offer mobile social applications with direct support for trajectories. The main contributions of this paper is the proposal of 25

8 interaction applications. In Proceedings of the 11th international conference on Web and Wireless Geographical Information Systems, W2GIS 12, pages 82 97, Berlin, Heidelberg, Springer-Verlag. [4] C. Cheng, F. Lu, and J. Cai. A quantitative scale-setting approach for building multi-scale spatial databases. Comput. Geosci., 35: , November [5] F. Giannotti, M. Nanni, F. Pinelli, and D. Pedreschi. Trajectory pattern mining. In Proceedings of the 13th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 07, pages , New York, NY, USA, ACM. [6] N. Meratnia and R. A. de By. Aggregation and comparison of trajectories. In Proceedings of the 10th ACM International Symposium on Advances in Geographic Information Systems, GIS 02, pages 49 54, New York, USA, ACM. [7] C. d. Mouza and P. Rigaux. Multi-scale classification of moving objects trajectories. In Proceedings of the 16th International Conference on Scientific and Statistical Database Management, pages 307, Washington, DC, USA, IEEE Computer Society. [8] J.-C. Muller and J.-P. Lagrange. GIS And Generalisation: Methodology And Practice (Gisdata, No 1). CRC Press, 1 edition, apr [9] MySQL Entrerprise. MySQL Workbench, Jun [10] S. Spaccapietra, C. Parent, M. L. Damiani, J. A. de Macedo, F. Porto, and C. Vangenot. A conceptual view on trajectories. Data Knowl. Eng., 65: , April [11] T. H. N. Vu, K. H. Ryu, and N. Park. A method for predicting future location of mobile user for location-based services system. Comput. Ind. Eng., 57:91 105, August [12] World Wide Web Consortium. W3c points of interest working group charter, Jun [13] J. J.-C. Ying, E. H.-C. Lu, W.-C. Lee, T.-C. Weng, and V. S. Tseng. Mining user similarity from semantic trajectories. In Proceedings of the 2nd ACM SIGSPATIAL International Workshop on Location Based Social Networks, LBSN 10, pages 19 26, New York, NY, USA, ACM. [14] H. Yoon, Y. Zheng, X. Xie, and W. Woo. Smart itinerary recommendation based on user-generated gps trajectories. In Proceedings of the 7th International Conference on Ubiquitous Intelligence and Computing, UIC 10, pages 19 34, Berlin, Heidelberg, Springer-Verlag. [15] Y. Zheng, X. Xie, and W.-Y. Ma. Geolife: A collaborative social networking service among user, location and trajectory. IEEE Data Eng. Bull., 33(2):32 39, [16] Y. Zheng, L. Zhang, X. Xie, and W.-Y. Ma. Mining interesting locations and travel sequences from gps trajectories. In Proceedings of the 18th International Conference on World Wide Web, WWW 09, pages , New York, NY, USA, ACM. [17] Y. Zheng and X. Zhou. Computing with spatial trajectories. Springer-Verlag New York Inc, [18] S. Zhou and C. B. Jones. Design and implementation of multi-scale databases. In Proceedings of the 7th International Symposium on Advances in Spatial and Temporal Databases, SSTD 01, pages , London, UK, UK, Springer-Verlag. 26