Development of a semantic recommendation system for auteur films


  Information Retrieval and   Machine Learning
  Prof. Dr.-Ing. Sahin Albayrak
  Dr.-Ing. Andreas Lommatzsch
  Partner: EYZ Media GmbH



The widespread availability of high speed, broadband Internet has spurred rapid growth in Video-On-Demand and in online video subscription services. To optimally support users in finding the most interesting movies from a complex collection of content, a recommendation system that evaluates both content-based knowledge and collaborative data is being developed that will suggest potentially relevant movies. The system also explains to users the reasons behind the offered recommendations, thus improving customer acceptance.


The project focuses on film recommendations for art house audiences. In addition to the traditional interpretation of repertory cinema, the claim 'Art House on Demand' addresses wider cultural contexts and includes 'straight-to-digital' as well as 'live' segments. The goal of the recommendation engine is to provide cross-device users with extensive support for discovering interesting movies and other content. The key challenges of the project lie in the sophisticated demands of this particular target group, the handling of multilingual content, the demands of cross-platform display solutions, and the heterogeneity of aggregated content sources.


We are developing and optimizing graph-based algorithms which link information from a range of sources and domains and offer a unified representation of knowledge in large graphs. The analysis and semantic enrichment of data aggregated from different knowledge sources (which also take user context into account) form the foundation of high-quality recommendations. To improve user acceptance of recommended films, the algorithms also explain the reasons behind the offered recommendations.