Rumormongers spread negative information throughout the social network, which may even lead to panic or unrest. Rumor should be blocked by spreading positive information from several protector nodes in the network. Users will not be influenced if they receive the positive information ahead of negative one. In many cases, network manager or government may not know the exact positions where rumor will start. Meanwhile, protector nodes also need to be selected in order to prepare for rumor blocking. Given a social network G = (V,E,P), where P is the weight function on edge set E, P_(u,v) is the probability that v is activated by u after u is activated. Assume there will be l rumormongers in the network while the exact positions are not clear, Robust Rumor Blocking(RRB) problem is to select k nodes as protector such that the expected eventually influenced users by rumor is minimized. RRB will be proved to be NP-hard and the objective function is neither submodular nor supermodular. We present an estimation process for the objective function of RRB based on Reverse Reachable Set(RR-Set) methods. A randomized greedy algorithm is designed for solving this problem. And this algorithm is proved to have approximation ratio \(\frac {1}{\alpha }(1-e^{-\alpha \gamma })(1+\epsilon )\) plus a constant, where γ is submodularity ratio and α is curvature.Finally, we evaluate our algorithm on real world data sets and do comparison among different strategies for protector. The results show the effectiveness and the efficiency of the proposed algorithm.

谣言传播者在整个社交网络中传播负面信息，这甚至可能导致恐慌或动荡。应该通过传播来自网络中多个保护节点的正面信息来阻止谣言。如果用户在负面信息之前收到正面信息，则不会受到影响。在许多情况下，网络管理员或政府可能不知道谣言的确切位置。同时，还需要选择保护者节点以准备阻止谣言。给定一个社交网络G =（V，E，P），其中P是边缘集E，P _（u，v）上的权重函数是激活u后u激活v的概率。假设在确切位置尚不清楚的情况下，网络中将有l个谣言贩子，那么稳健的谣言阻止（RRB）问题就是选择k个节点作为保护者，从而使谣言最终对最终用户的影响最小。RRB将被证明是NP-hard，目标函数既不是亚模也不是超模。我们提出了一种基于反向可到达集（RR-Set）方法的RRB目标函数的估计过程。为了解决这个问题，设计了一个随机贪婪算法。并且证明该算法具有近似比率\（\ frac {1} {\ alpha}（1-e ^ {-\ alpha \ gamma}）（1+ \ epsilon）\）加上一个常数，其中γ是次模数比，α是曲率。最后，我们在现实世界的数据集上评估我们的算法，并在保护器的不同策略之间进行比较。实验结果表明了该算法的有效性和有效性。