Abstract: Overlapping community detection is an essential topic in attribute networks, which aims to reveal the hidden structure of networks. However, most existing methods cannot effectively utilize both the topological and attribute information in the networks. Furthermore, these methods focus solely on reconstructing the local edge structure while neglecting the overall community structure. A modularity optimized graph convolutional autoencoder model OGCAE(Optimal Graph Convolutional AutoEncoder) for overlapping community detection is presented. Firstly, OGCAE employs deep convolutional neural networks for representation learning. Secondly, a Bernoulli-Poisson graph generation model for graph reconstruction is utilized. Finally, a contrastive modularity loss is incorporated into the loss function to obtain more discriminative features, which are beneficial for overlapping community detection. These three modules are jointly correlated through a unified loss function, which drives the parameter updates of the autoencoder and enhances the performance of overlapping community detection tasks. Experimental results on multiple real-world attributed networks demonstrate that OGCAE outperforms current state-of-the-art methods in terms of both accuracy and stability.