Identification of Source of Rumors in Social Networks with Incomplete Information (1509.00557v1)

Abstract: Rumor source identification in large social networks has received significant attention lately. Most recent works deal with the scale of the problem by observing a subset of the nodes in the network, called sensors, to estimate the source. This paper addresses the problem of locating the source of a rumor in large social networks where some of these sensor nodes have failed. We estimate the missing information about the sensors using doubly non-negative (DN) matrix completion and compressed sensing techniques. This is then used to identify the actual source by using a maximum likelihood estimator we developed earlier, on a large data set from Sina Weibo. Results indicate that the estimation techniques result in almost as good a performance of the ML estimator as for the network for which complete information is available. To the best of our knowledge, this is the first research work on source identification with incomplete information in social networks.