The recommender systems have been widely applied in numerous applications that support online retailers, video sharing websites, medical systems, etc. Similar measures are essential in providing valuable recommendations to users in such systems. This work presents a novel approach, namely Ball-Sim, with a new similarity metric using a balltree structure for recommender systems. Furthermore, we want to leverage the tree structure to determine the closest k nearby users to improve the recommender systems’ efficiency. The work’s experimental scenarios outlined the steps of building a balltree and identifying nearby users based on the tree structure. Besides, the work also evaluates the implemented recommender system by comparing the recommender system’s results based on the balltree-based spatial partitioning with the recommender system using the default parameters. The data used in the experiments is the Movielens dataset, a web-based film recommender system, and an important data source for evaluating the studies, with 100,000 samples, including ratings from 943 users for 1,664 movies. The results show that the recommender system with a balltree-based similarity metric can improve the accuracy compared to a commonly-used measure such as the cosine metric.