With the rapid development of location-based social networks, Point of Interest (POI) recommendation has become a research hotspot in the recommendation field. However, existing studies often fail to fully explore users’ POI interaction preferences and POI transition preferences, which limits the effectiveness of POI recommendations. This study proposes a personalized POI recommendation method based on a hybrid graph neural network. First, a user social network graph is constructed, and Graph Attention Networks (GAT) are employed to learn user social relationships. Second, a Signed Bipartite Graph is constructed between users and POIs, and user-POI interaction preferences are learned using Signed Bipartite Graph Neural Networks (SBGNN). Additionally, a directed POI transition graph is built, and POI transition preferences are learned using Session-Based Recommendation with Graph Neural Networks (SRGNN). Subsequently, user feature vectors incorporating social relationships and POI interaction preferences, as well as POI feature vectors that include user-POI interaction preferences and POI transition preferences, are generated. Finally, these feature representations are used as model inputs to recommend a Top-K POI sequence to users. Extensive experiments conducted on the Gowalla and Foursquare datasets demonstrate that the proposed method significantly improves both accuracy and recall compared to baseline models.