2013 International Conference on Computational Science
Recommendation algorithms have been researched extensively to help people deal with abundance of information. In recent years, the incorporation of multiple relevance criteria has attracted increased interest. Such multi-criteria recommendation approaches are researched as a paradigm for building intelligent systems that can be tailored to multiple interest indicators of end-users – such as combinations of implicit and explicit interest indicators in the form of ratings or ratings on multiple relevance dimensions. Nevertheless, evaluation of these recommendation techniques in the context of real-life applications still remains rather limited. Previous studies dealing with the evaluation of recommender systems have outlined that the performance of such algorithms is often dependent on the dataset – and indicate the importance of carrying out careful testing and parameterization. Especially when looking at large scale datasets, it becomes very difficult to deploy evaluation methods that may help in assessing the effect that different system components have to the overall design. In this paper, we study how layered evaluation can be applied for the case of a multi-criteria recommendation service that we plan to deploy for paper recommendation using the Mendeley dataset. The paper introduces layered evaluation and suggests two experiments that may help assess the components of the envisaged system separately.