Troeps documentation
Help for clustering
Clustering creates a (or several) viewpoint(s) from the set of
available objects in a concept. It is one of the most advanced
feature of the Troeps system.
For that purpose, Troeps will:
- Compute the dissimilarity between each pair of objects in the concept,
- Cluster the objects into a taxonomy of inclusive sets,
- Describe each set as a class in the new viewpoint.
It will be then up to the user to modify these viewpoints (changing
names, deleting classes or the whole viewpoint...).
It order to launch clustering, it must be provided with the
following information:
- The name of the new viewpoint which will be created and thus must
be different from any other viewpoint in the concept,
- The aggregation algorithm to be used for clustering (so far,
Troeps only implements a set of classical Sequential
Agglomerative Hierarchical Non-overlapping algorithms),
- The weight of the fields: it is necessary to weight the relative importance of
the fields in the computed similarity. This
is the role of the input to be given to the method. Each rank
must be a float value between 0. and 1. and ideally, their sum
must be 1..
Moreover, if a strictly positive weight is given to a field whose
type is another concept, Troeps will ask how to compute the
dissimilarity between the objects of that concept. It can use a
topological dissimilarity (basically computing the distance
between the classes to which the objects are attached) in an
already existing viewpoint or compute (through clustering) a new
viewpoint (to which in turn Troeps will ask for weights). This
last option is available by selecting "**new**".
For some algorithms, a set of parameters might be provided by the
user. In the current implementation, no parameter is available.
For more information, consult the Troeps
online documentation
Trpserver 1.3 -- Last revision 15/11/1999