In this paper we present a framework for learning models for Recommender Systems (RS) in the case where there are multiple implicit feedback associated to items. Based on a set of features, representing the dyads of users and items extracted from an implicit feedback collection, we propose a stochastic gradient descent algorithm that learn jointly classification, ranking and embeddings for users and items. Our experimental results on a subset of the collection used in the RecSys 2016 challenge for job recommendation show the effectiveness of our approach with respect to single task approaches and paves the way for future work in jointly learning models for multiple implicit feedback for RS.