Multicriteria predictors using aggregation functions based on item views

Multicriteria Collaborative Filtering is a promising approach to recommender systems that explores user ratings on item components in order to generate high quality recommendations. This paper focuses on multicriteria collaborative recommender systems and proposes a new algorithm that estimates aggregation functions, which represent the relative importance of individual components, based on the concept of item views. Experiments on a real multicriteria movie dataset demonstrate that our approach outperforms other aggregation models in terms of prediction precision and coverage. Furthermore, the study shows how the concept of item views (i) naturally emerges from the properties of the dataset, (ii) addresses the multicriteria recommendation problem, (iii) provides a mechanism to explain recommendations and (iv) drives the implementation of the rich user interfaces required by this type of recommender systems.