Community detection is an extremely important technology for today's rapidly evolving data mining and exploratory analysis. Many community detection algorithms have been proposed and applied. Among them, the label propagation algorithm (LPA) is widely used due to its closeness to linear time complexity and its simple characteristics. However, the results of detecting the community were found to be random when using LPA. Based on previous improvements to the LPA, this article proposes a new LPA (NOHLPA) that joins multilayer neighborhood overlap and historical label similarity. The NOHLPA considers both the node update order and label selection rules. We cited the label entropy as the basis for node update order and defined multilayer neighborhood overlap and historical label similarity for calculating node preferences to devise more appropriate label selection rules. We test the NOHLPA on five real datasets and compare NOHLPA with different combinations of algorithms. The experimental results show that the NOHLPA effectively improves the accuracy of community partitioning while ensuring stability.