<p>Identifying and analyzing overlapping community structures in complex networks holds significant theoretical importance and practical value. Due to the simplicity and efficiency, LPA-based (Label Propagation Algorithm) for detecting overlapping community structure has been widely used in complex networks. However, LPA-based algorithms are often caught in the dilemma of neighbor node random selection with low stability and accuracy only considering the network’s topology information but few prior knowledge or hindered by randomness. To address these issues, we propose an overlapping community detection algorithm, named MFLP, through node merging with a novel LPA strategy for large-scale networks. Specifically, MFLP first designs and employs a node merging strategy in which low-degree nodes are merged with their higher-degree neighbors to reduce the computational network scale. Then, the nodes are grouped and labeled according to the degree that starts with the lowest degree nodes and potentially diffuses the labels to their neighbors. Furthermore, a new node influence calculation method is introduced to improve the differentiation between pairs of nodes. Meanwhile, an overlapping node detection strategy is designed to identify overlapping nodes in the network according to their community affiliation. Extensive experimental results on both 10 real networks and 14 synthetic networks illustrate that our proposed algorithm MFLP achieves better accuracy and stability compared to 9 baseline models, especially suitable for overlapping community detection in large-scale networks.</p>