Database Reference
In-Depth Information
joining temporal and spatial information, greatly increases the search space of
most interesting problems, such as finding patterns or discovering discriminative
spatio-temporal features for classification or prediction problems.
One aspect of mobility data mining that the reader might have guessed by
reading this chapter is the fact that this research field still lacks an overall,
comprehensive, and clear theoretical framework. Such a framework should be
able to accommodate existing problems and solutions proposed in literature,
as well as clarify the relations between them. Some examples of efforts in
this direction exist in literature, and we also reported a few of them - for
instance, the relation between local trajectory patterns and global trajectory
classification models, and their abilities to grasp different, complementary kinds
of discriminatory features of trajectory data; or the relations between some of
the various forms of trajectory pattern. However, such cases are rather isolated,
and at the present, providing an integrated view of methods and issues is still a
largely unexplored part of the research field.
Another important point in mobility data mining is the fact that several
data sources might provide information about the same mobility phenomena
coming from different viewpoints. Each data source usually has distinctive
characteristics, strong points, and limitations, and their integration might help in
overcoming the limits of each of them. For instance, vehicle GPS data are usually
very detailed in space (i.e., spatial uncertainty is small) and time (frequency of
data acquisition is relatively high), yet it is inherently limited to the vehicles
that are involved in the data collection process; instead, mobile phone service
providers are able to collect information about mobility of all their customers,
and through the collaboration of a few providers it is possible to cover the
activity of very large portions of the real population. One example is
call detail
records (CDRs)
, which describe the cell towers that served each call performed
by each phone, together with the call's timestamp. CDRs allow us to build
mobility trajectories for each customer served. However, such trajectories are
very sparse (one point corresponds to a call, which are usually not so frequent)
and spatially rough (a point actually represents the whole area served by the
cell tower). Activities that try to combine these two data sources have begun
to appear recently, with the aim of improving the representativity of GPS data
through the extremely high penetration of the (spatially and temporally poor)
CDR data.
Finally, so far, our discussion has always implicitly assumed that the trajectory
data were analyzed offline and in a centralized setting, that is, by first collecting
all data in a single database and then analyzing them. However, mobility data are
usually massive and arrive as a continuous stream from the data source(s). Mas-
siveness and the streaming nature of data leads to make it impossible to collect
them, at a large scale, in a centralized database, and therefore analysis methods
