Mobility Data Mining - Mobility Data - page 106

6.1.2 Note on Terminology
In this chapter we will make frequent use of the term “trajectory pattern.”
As mentioned in Chapter 1 , the notion of trajectory pattern is substantially
equivalent to that of “trajectory behavior,” which also appeared in previous
chapters of this topic. The two notions originate from different communities and
simply reflect different perspectives of the same subject: the data management
view (where “trajectory behavior” originates) focuses more on determining
which trajectory is associated to each behavior; the data mining view, on the
contrary, is more focused on what are the interesting behaviors in the input
trajectories.
The several forms and variants of existing analysis tasks that belong to mobil-
ity data mining cannot be easily categorized into a set of fixed classes. However,
it is possible to recognize a few simple dimensions along which to locate the
different analysis methods. In the following we mention one of them, which will
also be used later as guideline during the presentation of analysis examples.
6.1.3 Local Patterns versus Global Models
The example of behavior illustrated at the beginning of this section is represen-
tative of a class of mining methods, called local patterns or, in most contexts,
simply patterns . The key point of local patterns is the aim of identifying behav-
iors and regularities that involve only a (potentially small) subset of trajectories,
and that describe only a (potentially small) part of each trajectory involved.
The complementary class of mining methods is called global models ,or
simply models . Their objective is to provide a general characterization of the
whole data set of trajectories, thus going toward the definition of general laws
that regulate the data, rather than spotting interesting yet isolated phenomena.
For instance, we will see later mining tasks aimed to define a global subdivision
of all trajectories into homogeneous groups, as well as tasks aimed to discover
rules able to predict the future evolution of a trajectory (i.e., the next locations
it will visit).
In the rest of the chapter we will provide an overview of the problems and
methods available in the mobility data mining field. For obvious reasons of
space, the discussion will not cover exhaustively the available literature on the
subject, and instead will propose some representative examples of the various
topics. The presentation will mainly follow the distinction between local patterns
and global models already introduced. In this chapter we will assume that raw
location information, such as GPS traces, has already been preprocessed to
obtain trajectories according to the discussions provided in Chapter 2 ,and
will not consider the additional issues related to uncertainty already tackled
in Chapter 5 . Besides the examples provided here, the reader can find some