Existing link prediction methods often describe the similarity of nodes to be predicted with common neighbors but ignore the contribution of many non-common neighbors and more than second-order non-global paths to the similarity of nodes. To solve this problem, a link prediction algorithm based on the connected coupling degree of nodes and the preference degree of topology connectivity is proposed. Starts with common-neighbor nodes, the algorithm reversely searches to the predicted nodes, builds a second-order connection including non-common neighbor nodes, and forms the sets of generalized common neighbors and the fifth order extended local path between prediction nodes. Extract the node attributes and edge information between nodes and design a computing method of connected coupling degree of nodes, which not only makes full use of the network local topology information, enhances the homogeneity of neighbor’s degrees, but also distinguishes the differences between neighbor nodes, to improve the accuracy of similarity calculation. In addition, the preference degree of topological connectivity is defined by the node’s numbers and paths’ length, which measures the closeness of the connection between nodes, describes the ability of nodes to promote the connection of nodes to be predicted, and improves the accuracy of prediction. Compared with 5 representative link prediction benchmark algorithms based on common neighbors (CN, AA, Car) and network structure (LP, Katz), ten real network simulation experiments show that the prediction accuracy of the algorithm in this paper is higher; Furthermore, lower computation complexity and higher execution efficiency is shown on this algorithm compared with the typical local path (LP) and global path (Katz) algorithm.