Recommendation systems are an effective tool for dealing with such issues; nevertheless, the Recommendation System suffers from several issues, including sparse data, cold-start, and inaccuracies. To tackle these issues, we design the Deep LocAWS model, a new Deep multilayer perceptron Neural Network (NN) Collaborative Filtering for location-aware web services recommendations. A Deep LocAWS structure is made up of three components: A Deep MLP Neural model (including input layer, MLP layers, and output layer). Deep LocAWS adds the following features: Characteristics of the users’ and services’ identifiers are first translated into N-dimensional dense embedded vectors, and after that Multi-Layer-Perceptron (MLP) accepts N-dimensional and non-linear features and applies a cosine-based similarity, adaptive corrector. We implemented an advanced Huber Loss function to improve Deep LocAWS's performance across all criteria. Deep LocAWS is not only capable of learning N-dimensional and non-linear connections between users’ identifiers and service identifiers, but this also significantly reduces cold-start, data sparseness, and inaccuracies problems. Deep LocAWS recommendations performance knocks previous location-aware services recommender systems in simulations based on the real-world WS dream dataset.