Data mining aims at extracting knowledge from large sets of data such as association rules, clusters and patterns. When both associations and temporal order between items are sought, the discovered knowledge are called sequential patterns. Existing studies were conducted mainly on sequential patterns involving objects and in some cases object categories. While patterns based on objects are too specific, non frequent patterns based on categories (concepts) may have different levels of abstraction and be possibly less precise. Taking into account a given domain ontology during a data mining process allows the discovery of more compact and relevant patterns than in case of the absence of such source of knowledge. Moreover, objects may not be only expressed by the concepts they are attached to, but also by the semantic links that hold between concepts. However, related studies that exploited domain knowledge are restrictive with regard to the expressive power offered by ontology. Our contribution consists to define the syntax and the semantics of a pattern lan- guage which exploits knowledge embedded in an ontology during the process of mining sequential patterns. The language offers a set of primitives for pattern description and manipulation. Our data mining technique explores the pattern space level by level using a set of navigation primitives which take into account the generalization/spécialization links that hold between concepts (and relationships) contained in patterns at different abstraction levels. In order to validate our approach and analyze the performance and scalability of the proposed algorithm, we developed the OntoMiner plateform. Throughout this thesis, the potential of our mining approach was illustrated with an ex- ample of Web recommendation. We came to the conclusion that taking into account con- cepts and relationships of an ontology during the process of data mining allows the dis- covery of more relevant patterns and leads to better recommendations than those found without using background knowledge.
