Overlapping community discovery based on graph embedding and label propagation algorithm

Traditional label propagation algorithms (LPA) exhibit instability and poor accuracy in community discovery, primarily due to random node selection, uncertain label update sequences, and neglect of node importance variations. We present GELPA-OCD (overlapping community discovery based on graph embedding and label propagation algorithm), an overlapping community discovery algorithm that integrates graph embedding with label propagation to address these limitations. Our approach introduces a multidimensional node importance assessment strategy and employs Node2vec graph embedding to represent nodes as low-dimensional vectors, effectively capturing network structure features. The algorithm employs similarity-based weight factors to guide label propagation and implements adaptive filtering mechanisms to enhance effectiveness. We conduct experiments on both real and artificial datasets. Using EQ, NMI , and F1-score as evaluation metrics, the experimental results show that the proposed algorithm effectively reduces randomness and uncertainty in node selection and label updating processes, achieving more stable and accurate overlapping community discovery.