Towards Semantic Trajectory Knowledge Discovery
Luis Otavio Alvares1;2, Vania Bogorny1, Bart Kuijpers1,
Bart Moelans1, Jose Antonio Fernandes de Macedo3, Andrey Tietbohl Palma2
1 Theoretical Computer Science,
Hasselt University & Transnational University of Limburg, Belgium
2Instituto de Informa´tica – UFRGS, Brazil
3Ecole Polytechnique Fe´de´rale de Lausanne, Switzerland
Abstract. Trajectory data play a fundamental role to an increasing number of
applications, such as transportation management, urban planning and tourism.
Trajectory data are normally available as sample points. However, for many ap-
plications, meaningful patterns cannot be extracted from sample points without
considering the background geographic information. In this paper we propose
a novel framework for semantic trajectory knowledge discovery. We propose
to integrate trajectory sample points to the geographic information which is
relevant to the application. Therefore, we extract the most important parts of
trajectories, which are stops and moves, before applying data mining methods.
Empirically we show the application and usability of our approach.
1. Introduction
Trajectory data are normally obtained from location-aware devices that capture the po-
sition of an object at a specific time interval. The collection of these kind of data is
becoming more common, and as a result large amounts of trajectory data are available in
the format of sample points. In many application domains, such as transportation manage-
ment, animal migration, and tourism, useful knowledge about moving behavior or moving
patterns can only be extracted from trajectories if the background geographic information
where trajectories are located is considered. Therefore, there is a necessity for a special
processing on trajectory data before applying data mining techniques.
An example which expresses such necessity is shown in Fig. 1. In Fig. 1 (left)
we can visualize a set of trajectories, that apparently have no meaning. In Fig. 1 (right)
we have the same trajectories over the geographic space, where we can visually infer the
geographic location (Paris) and the intersection of trajectories with touristic places (e.g.
Eiffel tower) and hotels.
Raw trajectory data are collected from the Earth’s surface similarly to any kind of
geographic data (see raw data level in Fig. 2). It is known that raw geographic data require
a lot of work to be transformed into maps, normally stored into shape files (processed
data level in Fig. 2). The processed geographic data are generated for any application
domain, and therefore are application independent. They can be used, for instance, to
build geographic databases of applications of transportation management, tourism, ur-
ban planning, etc (application level in Fig. 2). Geographic databases, on the other hand,
are application dependent, and therefore, will contain only the entities of processed geo-
graphic data that are relevant to the application.

Figure 1. (left) trajectories and (right) trajectories with geographic information
Data mining and knowledge discovery are on the top level, and are also application
dependent. In any data mining task the user is interested in patterns about a specific
problem or application. For instance, transportation managers are interested in patterns
about traffic jams, crowded roads, accidents, etc, but are not interested in, for instance,
patterns of animal migration.
In trajectory pattern mining, however, mining has been basically applied over raw
trajectory data or trajectory sample points, as shown in Fig. 2. Indeed, these algorithms
as far as we know have not considered the background geographic information in the
mining process. We claim that meaningful patterns for decision making processes in real
applications cannot be extracted from raw trajectory data. For data mining and knowledge
discovery, trajectories represented as sample points need to be a priori integrated with the
background geographic information which is relevant to the application, for then apply
data mining methods. This a priori integration will generate semantic trajectory patterns,
that facilitate the user’s task to analyze and interpret the knowledge in post-processing
steps.
In this paper, we propose a general framework to integrate geographic data with
trajectories in the form of sample points, in order to create semantic trajectories for knowl-
edge discovery. Therefore, we will consider trajectories as a set of stops and moves, where
stops are the important places for the application, defined by the user, and moves are tran-
sitions between consecutive stops.
The remaining of the paper is organized as follows: Sect. 2 presents the related
works and the main contributions. In Sect. 3 we introduce the basic concepts about trajec-
tories, stops, and moves. In Sect. 4 we present a framework to integrate geographic data
with trajectories in order to create a semantic trajectory database for knowledge discovery.
In Sect. 5 we present some application examples to show the usability of our approach.
In Sect. 6 we conclude the paper and present some directions of future work.
2. Related Work and Contribution
Several data models have been proposed for efficiently querying raw trajectory
data such as [Kuijpers and Othman 2007, Wolfson et al. 1998], but only a few ap-
proaches [Brakatsoulas et al. 2004, Spaccapietra et al. 2007, Mouza and Rigaux 2005]