Learning Semantic Recommenders

Competence Center:  Information Retrieval and Machine Learning
ContactProf. Dr.-Ing. Sahin AlbayrakDr.-Ing. Andreas Lommatzsch
Website: --


The goal of this project is to evaluate how recommender systems can be automatically adapted to semantic data sets.

Recommender systems are algorithms that compute a ranked list of objects for a user in a given context. The complexity of the data sets upon which typical recommender systems work continuously increases with the growth of the Internet and the higher interleaving and semantic enrichment of web-based services. To present a user with high-quality recommendations, the recommender system has more and more incorporate knowledge about the objects in the data set. This increase in complexity, which is reflected both in the higher diversity of object types as well as the richer set of object relationships, is not or not adequately addressed in current recommender systems.

Typically, these systems are designed to handle a static set of object types and their relationships, and are not adaptable to changes in both. The goal of this project is to develop and evaluate learning strategies for semantic recommendation algorithms, in order to enable these algorithms to automatically recognize complex semantic relationships in the underlying data set, and thus to improve the quality of recommendations.

The developed algorithms will be prototypical implemented in the Semantic Engine: implemented.