Recommender Systems (RS) have become very imperative in several fields such as e-commerce, and social media networking. In recommender systems, there is a problem of filtering information, which is considered one of the complex challenges in building these systems. Recently, there are many algorithms available to address RS’s challenges, one of the most common algorithms is collaborative filtering recommendation. This algorithm is based on Alternating Least Squares (ALS) is one of the widespread algorithms using matrix factorization method of recommendation system which is using to address that intricate challenges. In this paper, we suggest an approach called improved ALS to improve the performance of conventional ALS model on two different datasets, Movielene and Book-crossing using Apache Spark. The model evaluation is done using Root Mean Squared Error (RMSE) metrics, as well as compared with the conventional ALS Model.