Present the concept of candidate node pair (CNP) and its characteristics. We perform the link prediction only within the CNPs so as to reduce the computation time.

Define the patterns covered by the CNPs and their weights. The final prediction score of a CNP can be calculated using the weights of the patterns it covers.

Present a method to efficiently calculate the prediction scores of the CNPs without explicitly enumerating all the CNPs.

Present an algorithm for link prediction in bipartite network. Its time complexity is O(m), while those of other algorithms are O(mn) or O((m + n)³).

Experimental results show that the methods presented can obtain higher precision results in less time than other methods.