Abstract: Classifying data hierarchically is an efficient approach
to analyze data. Data is usually classified into multiple categories, or
annotated with a set of labels. To analyze multi-labeled data, such
data must be specified by giving a set of labels as a semantic range.
There are some certain purposes to analyze data. This paper shows
which multi-labeled data should be the target to be analyzed for
those purposes, and discusses the role of a label against a set of
labels by investigating the change when a label is added to the set of
labels. These discussions give the methods for the advanced analysis
of multi-labeled data, which are based on the role of a label against
a semantic range.
Abstract: Data objects are usually organized hierarchically, and
the relations between them are analyzed based on a corresponding
concept hierarchy. The relation between data objects, for example how
similar they are, are usually analyzed based on the conceptual distance
in the hierarchy. If a node is an ancestor of another node, it is enough
to analyze how close they are by calculating the distance vertically.
However, if there is not such relation between two nodes, the vertical
distance cannot express their relation explicitly. This paper tries to fill
this gap by improving the analysis method for data objects based on
hierarchy. The contributions of this paper include: (1) proposing an
improved method to evaluate the vertical distance between concepts;
(2) defining the concept horizontal distance and a method to calculate
the horizontal distance; and (3) discussing the methods to confine a
range by the horizontal distance and the vertical distance, and
evaluating the relation between concepts.