The complete extraction of local patterns from binary relations has
been studied extensively. Among others, it concerns the computation
of every closed sets or formal concepts. It can be applied across
many application domains, for instance, basket data analysis,
community discovery on the WWW, but also gene regulation
understanding. We now study the systematic extension of available
methods to n-ary relations. This includes the concept of closed patterns
in n-ary relations, properties of user-defined constraint and the
design of associated solvers, extensions towards fault-tolerance,
the multiple use of such patterns to support various data mining
tasks (e.g., feature construction, local pattern discovery, clustering).
This thesis starts from the availability of complete collections of
closed patterns that hold real-life n-ary relations. Innovative
post-processing strategies have to be designed to support knowledge
discovery processes. We will have targeted applications in dynamic
network mining and this will be a new appealing context to develop
further the inductive database framework.