Database Reference
In-Depth Information
we can understand when the patterns remain similar, when and how they change,
or when they disappear.
Pattern Interpretation
The intrinsic difficulty of behavior extraction lies in the need of integrating into
the discovery process the contextual knowledge. We define contextual knowl-
edge any kind of information that is not only related to the geometric parts of a
trajectory and that has some relation with the mobility data. Examples of con-
textual knowledge are: the geographical environment where the objects move
(e.g., hotels, roads, parks), any nongeometric moving object feature (e.g., the age
of the tracked person), or the application-specific concepts and behavior (e.g.,
goal of the movement or predefined behavior, such as commuting, shopping, or
touring).
Application domain knowledge may be globally represented by formally
encoding it into a knowledge representation structure such as an ontology ,which
can be used to represent the main concepts of the application. Formal ontolo-
gies are described by languages that are formal and machine readable. They
often include reasoning facilities that support the automatic processing of that
knowledge. Standards such as description logics (DL) provide a deductive infer-
ence system based on a formal, well-founded semantics. The basic components
of DL are suitable to represent concepts, properties, and instances. Complex
expressions, called axioms , can be used to implicitly define new concepts. Com-
bining ontologies with data mining is an intricate, challenging, and growing
research field. Besides, in the case of mobility, additional difficulties due to the
complexity of the managed data and patterns make this combination even more
arduous. The lack of primitive spatio-temporal ontology representation and rea-
soning mechanisms is the major obstacle for the successful development of this
trend.
Some recent proposals are making the first steps in this direction involving
contextual knowledge in the form of ontologies. An interesting feature of com-
bining data mining with ontologies in the knowledge discovery process is the
possibility of integrating deduction and induction aspects. The inductive power
of the data mining, extracting patterns from data (bottom-up), is enriched with
the possibility to deductively infer additional information based on some appli-
cation domain knowledge (top-down). This combination allows us to classify
the mobility patterns, as extracted from the mining step, into the application
knowledge concepts encoded in the ontology. An example of this induction-
deduction combination is the framework Athena, an extension of M-Atlas that
is an attempt to exploit ontologies in the mobility knowledge discovery process.
Athena represents application domain knowledge in an ontology where axioms
define the behavior we want to find in the data. Therefore a classification of the
extracted pattern into predefined behavior is performed directly by the ontology
Search WWH ::




Custom Search