Recommendation systems were designed to solve the problem of data overload. The objective is therefore to select from a large number of items those of low quantity relevant to a given user. Taking into account the repetitive and periodic nature of interactions between users and items has improved the performance of existing systems. But these systems do not take into account the digital data associated with these interactions. In this paper, we propose a recommendation approach based on gradual patterns which makes it possible to model the covariations between items. The experimental results obtained with proposed approach are encoraging with respect to the used dataset.