Many patterns have been uncovered in complex systems through the applicationof concepts and methodologies of complex networks. Unfortunately, the validityand accuracy of the unveiled patterns are strongly dependent on the amount ofunavoidable noise pervading the data, such as the presence of homonymousindividuals in social networks. In the current paper, we investigate theproblem of name disambiguation in collaborative networks, a task that plays afundamental role on a myriad of scientific contexts. In special, we use anunsupervised technique which relies on a particle competition mechanism in anetworked environment to detect the clusters. It has been shown that, in thiskind of environment, the learning process can be improved because the networkrepresentation of data can capture topological features of the input data set.Specifically, in the proposed disambiguating model, a set of particles israndomly spawned into the nodes constituting the network. As time progresses,the particles employ a movement strategy composed of a probabilistic convexmixture of random and preferential walking policies. In the former, the walkingrule exclusively depends on the topology of the network and is responsible forthe exploratory behavior of the particles. In the latter, the walking ruledepends both on the topology and the domination levels that the particlesimpose on the neighboring nodes. This type of behavior compels the particles toperform a defensive strategy, because it will force them to revisit nodes thatare already dominated by them, rather than exploring rival territories.Computer simulations conducted on the networks extracted from the arXivrepository of preprint papers and also from other databases reveal theeffectiveness of the model, which turned out to be more accurate thantraditional clustering methods.
展开▼
