La Recommandation Des Localisations : Une Nouvelle Approche Basée Sur La Fouille Incrémentale Des Règles Séquentielles

Résumé: In recent years, the rapid development in the technological field has stimulated the development of location acquisition and mobile communication technologies. This development has created many location services, such as the recommendation of points of interest (POI). POI recommendation consists of suggesting places that a user might be interested in visiting. The purpose of this service is to help mobile users to discover new and interesting places (for example, restaurants and shops), while on the move. In the literature, many models for recommending POIs have been proposed. These models take into account various factors such as geographic, temporal and social influences. Although these models have proven to be efficient, few of them take into account the sequential behavior of human mobility. Additionally, several models have been designed with static data in mind, thus ignoring the continuous recording and collection of check-in data in the LBSN (Location Based Social Network). This has led to the design of recommendations that must be formed from scratch to make up-to-date predictions, when new check-in data arrives. The temporal and spatial complexity of these recommenders can therefore increase considerably when applied to dynamic data. Therefore, there is a need to design incremental recommenders capable of handling dynamic data efficiently. To accommodate these limitations, this thesis proposes a new point of interest recommendation approach, called STS-Rec. The latter is mainly based on the extraction of sequential rules and it takes into account the sequential behavior of human mobility. STS-Rec first transforms mobility data into location sequences. Then, it gradually extracts sequential recommendation rules from these sequences. An experimental evaluation conducted on large-scale real-life check-in data from Brightkite shows that the proposed model outperforms the static version of the system in terms of execution time and space.

Publié dans la revue: