Hybrid recommender systems combine different approaches to provide better recommendations. The most common hybrid algorithms mix collaborative, content-based, demographic filtering among others. However, these hybrid approaches seldom consider the user-recommender interaction. In this paper, we propose a new hybrid recommender system through considering the user-recommender interaction. First, we define the recommender and user behaviors. The recommender system accepts user request, recommends N items to the user and records user choice. Second, we employ the recall metric to evaluate the quality of the recommender. The number of recommendations in each turn essentially serves as the accuracy constraint. Third, we test the random, kNN and our hybrid algorithm with the new metric. Specifically, we study the impact of different granules to the performance of our algorithm. Experiments results on the well-known MovieLens dataset show that the hybrid algorithm performs better, and appropriate granule selection is essential.