This thesis gives a generalization of the recommendation principle of recommender systems. Instead of considering a recommendation as an item, it is considered as a combination of several items following a given pattern. A recommendation of a unique item is then a particular case of this type of recommendation. The proposed recommender system framework is based on an architecture derived from work in adaptive hypermedia systems. Three layers are defined: a semantic layer, a user layer and an intelligence layer. The semantic layer consists of two sub-layers, a sub-layer modeling the content according to the general knowledge of the domain and a sub-layer modeling the content according to the specific knowledge of the application. This second sub-layer allows taking into account the expertise of the offer’s supplier for the proposals of the recommender system. The user layer models the user into the recommender system and the intelligence layer contains recommendation process. We propose to decompose the adaptation into two main processes, a process called projection of domain’s individuals on user profiles and a combinatorial research process. The first brings a weight, called the note in most recommender systems, giving probable users’ interests on the different items. The combinatorial research process searches among the many possible combinations a suitable solution (optimal if possible) to propose to the user. This framework of combinatory recommender systems is applied to the domain of tourism for the company Côte-d'Or Tourisme implied in the research contract. The purpose of this application is to offer the user a set of tourism offers in the form of journey. This tourism problem leads to the formal definition of a combinatory optimization problem which is a variant of the knapsack problem. To resolve this kind of problem, it is necessary to use a metaheuristic to tend toward a good solution in a reasonable time. We present an algorithm based on simulated annealing and a multi-objective algorithm to solve this problem. The instantiation of each layer of the framework for the tourism system is described in detail. Finally, this thesis presents a mobile application serving as a user with the tourism recommender system and it presents the technical developments for this project.
