针对二分网络图中基于链路预测的ILP算法在投影过程中只采用了底部节点,从而导致网络拓扑信息利用不充分的问题,文章提出改进的ILPExt算法,将顶部投影信息和底部投影信息结合起来,根据二部图的底部节点和顶部节点分别生成底部投影图和顶部投影图,然后在2个投影图中分别找出顶部节点和底部节点的内部链边集,并进行合并,得到预测链路的边集。实验结果表明,改进的ILPExt算法比ILP算法能够找出更多的链路预测边,并且提高了召回率。
The network topology information is underutilized because the ILP algorithm based on bipar- tite network link prediction uses only the bottom nodes in the projection process. An improved algo- rithm ILPExt that adopts the projection information of the top and bottom nodes is proposed in this paper. Firstly, the top and bottom projection graphs are generated by the top and bottom nodes of the bipartite network. Secondly, the internal link sets of the top and bottom nodes are found in the two projection graphs respectively. Then the two sets of internal links are merged to obtain the total link prediction edges. The experimental results show that the proposed ILPExt algorithm gets more link prediction edges and improves the recalls, which outperforms the II.P ~ltrori*hrn