In this paper, authors mainly focus on representing users and movies as vectors in a better way. This step is essential for a recommendation system, which determines the performance of the recommendation system. Through the Deep Learning technique, by combining the traditional Collaborative Filtering algorithm with matrix factorization algorithm, we propose a new model architecture based on the NeuralCF model\cite{he2017neural} and Youtube Two Tower model, which we call the "Two Tower NeuralCF model". The model was tested on the Movielens 100k and 1M datasets and obtained better results than the NeuralCF model.