Database Reference
In-Depth Information
the analysis are still valid in the real world. Other techniques aimed at getting
useful results from the mining step include progressive clustering and parameter
tuning. The second part of the chapter focuses on trajectory behavior as already
introduced in Chapter 1, distinguishing between spatio-temporal and semantic
behavior. We illustrate how to extract semantic behavior such as “StuckInTraffic-
Jam” or “Commuter” using a semantic-enriched mobility knowledge discovery
process.
7.2 The M-Atlas System
M-Atlas 1 is a running system developed to handle all the steps of the mobility
knowledge discovery process. M-Atlas is a querying and mining system based
on extensions of SQL and centered on the concept of trajectory. Besides the
mechanisms for storing and querying trajectory data, M-Atlas has mechanisms
for mining trajectory patterns and models that, in turn, can be stored and queried.
The basic design choice is compositionality , that is, querying and mining of tra-
jectory data; patterns and models may be freely combined in order to provide the
expressive power needed to master the complexity of the mobility knowledge
discovery process. The conceptual model behind M-Atlas views the knowledge
discovery process as the interaction between two conceptual worlds: the data
world and the model world. The former is a set of entities to be mined, tra-
jectories in our case; the latter is a set of models and patterns extracted from
the data, representing the result of mining tasks. Two kinds of operators con-
nect the two worlds: the mining operators, and the entailment operators. Mining
operators map data into models, or patterns, while entailment operators map
models, patterns, and data into the data that satisfy the property expressed in
the given model or pattern. This view supports compositionality, as data can
be mapped onto models and vice versa, coherently with inductive database
vision. Another design choice of the system is that all entities are represented
in the object-relational data model, which is more suitable to tackling the struc-
tural complexity of spatio-temporal data compared with the standard tabular
data.
The M-Atlas system is equipped with a graphical user interface and a set
of interactive graphical tools allowing the user to navigate the data and model
easily. This has the advantage of making the tool usable by domain expert users
to get full advantage of their domain expertise. Each interaction of the analyst
with the interface is compiled into a sequence of M-Atlas queries that can be
retrieved at any moment to describe or review the entire process. Alternatively,
an expert data mining analyst can directly submit queries to the M-Atlas engine,
to exploit its full expressiveness.
1 Available for download at the address http://www.m-atlas.eu
Search WWH ::




Custom Search