--

The recent success of neural networks has boosted research on pattern recognition and data mining. Machine learning tasks, like object detection, machine translation, and speech recognition, have been given new life with end-to-end deep learning paradigms. This work deals with the field of Graph Neural Networks (GNNs) and Spatial Graph Neural Networks (SGNNs), with a special focus on their applications in recommenda- tion systems. The aim is to provide a comprehensive overview of the topic through a series of thematic sections. We begin our journey by examining the evolution of GNNs and SGNNs, highlighting models that have demonstrated excellence in recommendation systems. The key dataset used for the research, was obtained from Yelp and than, was rigorously filtered to include only data related to the Philadelphia area in the United States. This geographical restriction provides a focused perspective on business activity and reviews within this specific region. The Spatial Graph Message Passing (SGMP) model by Zhang and Liao was chosen as reference for our study. Here, the main features of theis model are discussed, highlighting strengths and limi- tations, based on existing literature and our own analysis. Additionally, a description of the methods used to process the data is provived, and, furthermore, the SGMP model is validate by using authors’ settings and by modifying some parameters (such as training epochs, batch size, learning rate) to seek performance improvements. Subsequently, we present the results of our experiments on Yelp data, evaluating the accuracy of the GNN’benchmark model by Jan Eric Lenssen and Matthias Fey, using a dropout-type regularisation and performing fine-tuning to maximise performance. In addition, we propose a possible modification of the Zhang and Liao model to enable its use in link prediction tasks through node classification rather than graph classification. This study has highlighted the potential of SGNNs in the field of recommendation systems and mobility modelling. With continued research and refinement, these mod- els hold the promise of making a significant contribution to the delivery of tailored recommendations and the advancement of our understanding of human mobility pat- terns. We argue that this research can provide a foundational framework for future investigation and application of SGNNs in a wide range of domains. i