Projection-based link prediction in a bipartite network

Abstract

An algorithm for link prediction in a bipartite network is presented. In the algorithm, we first map the bipartite network onto a unipartite network called a projected graph. Based on the projected graph, we define the concept of a candidate node pair (CNP). We perform link prediction only within the CNPs to reduce the computation time. We also define the patterns covered by the CNPs and weights of the patterns. By calculating the weights of the patterns that a CNP covers, the connectivity of the CNP can be obtained, which can be used as the final score of link prediction. For a bipartite network with n and m nodes in the two parts, the time complexity of the proposed algorithm is O(m), whereas those of other algorithms are O(mn) or O((m+n)3). The experimental results show that our algorithm can achieve higher speed and superior quality link prediction results in bipartite networks compared with other methods.