In recent years, the ubiquitous GPS-equipped mobile devices have substantially advanced the research on personalized user mobility patterns through location data. However, accurately predicting Points of Interest (POIs) remains challenging in scenarios with sparse data. This study addresses such challenges by employing a self-supervised contrastive learning approach, which utilizes the adjacency matrices of locations within cities to predict users' subsequent POIs. This method not only improves prediction accuracy where data is limited but also sets the stage for innovative applications in deciphering human mobility and enhancing personalized services.