Recommender Systems

Contact: Benjamin KilleAndreas Lommatzsch


The "Recommender Systems"-Cluster is focused on finding the correct items for the people at the correct time. Recommender systems use any available information to predict what is and will be interesting for the users. Often, information such as what other people have seen and liked (usage data), or the item content and user profiles, is used to generate recommendations. 

Semantic Recommender: Semantic recommenders are recommenders that use semantic technologies to implement complex hybrid recommender systems. Due to the complexity of data available to recommenders, a recent trend has emerged of modeling this data as a semantic network, and applying semantic technologies to implement recommender systems. Semantic approaches view all data as a repository of semantic triples, allowing generic machine learning and information retrieval approaches to be used. Furthermore, generic repositories of semantic triples can be integrated into semantic recommendation systems to enrich the available knowledge, and provide better recommendations to the user.

Context-Aware Recommender: Most of todays recommender systems produce static recommendations, not taking into account parameters such as time of day or year, whether or not the user is in the company of others etc., instead they are based on item and user similarities. Finding context-dependent features and feature-dependencies in usage data combined with data about the surroundings (company, time, weather, etc.) allows for more accurate recommendations. Context-aware recommendations can be expressed as ad hoc recommendations, in contrast to the more static recommendations of standard recommender systems.

Evaluation of Recommender Systems: One of the main challenges in the development of recommender algorithms is the appropriate evaluation of the performance of different approaches. Since recommender systems focus around the user, standard evaluation metrics which are common in the information retrieval domain cannot easily adopted for evaluation.