Abstarct: Recommender systems, which are systems to help users in choosing favourite resources, have been emerged recently, and are being developed rapidly. These systems provide a customized environment via incorporating the past interactions of users and recognizing their interest to choose appropriate resources. Modeling user’s behavior and the mechanism of presenting a recommendation are considered as the main issues in developing recommender systems. Existing methods utilize users’ interests as a latent factor in modeling their behavior. The “interest” represents a set of resources with the same characteristics that is chosen by the user in the past. It is inferred from the user’s profile. Ignoring the dynamic of a user’s interests and missing to prioritize between the different interests in various times are considered as the main challenges in the current methods. In this research, the “preference”, which shows the priority of an interest in a certain time, is considered along with the “interest” to model a user’s behavior. The number, time and score of the chosen resources are the main factors in computing a user’s preferences. In the proposed method, the behavior of each user is modeled by using the paired ≺interest,preference≻. We utilize the user’s profile in a Bayesian network frxamework named as Dirichlet Process Mixture Model to create user model. Then, similar (neighbor) users are gathered baxsed on the generated behavior model to provide recommendations. Dynamism of the number of interests and adaptive preferences are the main contribution of the proposed method. Also, the proposed model has one hyper parameter which is initialized heuristically. We used a news and movie recommender system as a case study to evaluate the proposed model. The news dataset was collected from Twitter (news channels). Also, the MovieLens was employed as a movie databaxse. The performance of the proposed model was compared with the state-of-the-art methods such as TimeSVD, Content-baxsed, non-Negative Matrix Factorization, Item-Item Collaborative Filtering. The results of the comparison indicate the efficiency of the proposed model.