Bayesian nonparametric (BNP) methods have become very popular over recent years in machine learning and statistics as it allows to build elegant and sophisticated models. Contrary to Bayesian parametric methods, this set of techniques allows the number of parameters to grow with the number of data and is particularly suitable in the data rich environment we now face. This project aims at developing new Bayesian models for the probabilistic modeling of large and structured data such as networks and buyer preferences.First, we aim at developing new models for networked data. The last few years have seen a tremendous interest in the study and understanding of complex networks. We plan to develop new models for static and dynamic networks, with or without clustering structure, that can handle a potentially large number of nodes and exhibit a power-law behavior, with simple inference procedures for the parameters.Second, we aim at developing BNP recommender systems. Recommender systems aim at predicting the preference that a user would give to a specific item. They are especially useful for e-commerce in order to provide targeted advertisements to users. When the number of potential users and items is potentially large compared to the number of transactions, a BNP approach becomes sensible. We aim at developing new probabilistic models for the modeling of the behavior of buyers over time.