Abstract
Link prediction in complex networks aims at predicting the missing links from available datasets which are always incomplete and subject to interfering noises. To obtain high prediction accuracy one should try to complete the missing information and at the same time eliminate the interfering noise from the datasets. Given that the global topological information of the networks can be exploited by the adjacent matrix, the missing information can be completed by generalizing the observed structure according to some consistency rule, and the noise can be eliminated by some proper decomposition techniques. Recently, two related works have been done that focused on each of the individual aspect and obtained satisfactory performances. Motivated by their complementary nature, here we proposed a new link prediction method that combines them together. Moreover, by extracting the symmetric part of the adjacent matrix, we also generalized the original perturbation method and extended our new method to weighted directed networks. Experimental studies on real networks from disparate fields indicate that the prediction accuracy of our method was considerably improved compared with either of the individual method as well as some other typical local indices.
Highlights
Link prediction aims at revealing missing or potential relations between data entries from large volumes of data sets which are subject to dynamical changes and uncertainty
In the context of complex networks which are usually described as graphs, the global topological information always lies in their adjacency matrices, in which nonzero entries denote links between corresponding nodes, while zero ones denote missing or nonexistent links
Following the idea that the consistency in network structure can be represented by the eigenvectors of its adjacent matrices, the authors[10] proposed a structural perturbation method (SPM) in which a new matrix was constructed for prediction by perturbing the eigenvalues of the adjacent matrix while fixing the eigenvectors
Summary
Link prediction aims at revealing missing or potential relations between data entries from large volumes of data sets which are subject to dynamical changes and uncertainty. The CN index[11] of a node pair is the inner product of their corresponding rows of the adjacent matrix, and the RA index[11] of some weighted adjacent matrix whose column sum is assigned as 1 These are local indices that have explicit physical meanings, while the Katz index[12] uses the global information obtained from some series of the adjacent matrices. Motivated by these observations, new link prediction methods based on different kinds of manipulations of the adjacent matrices have been developed. In14, by introducing the robust PCA technique, the authors developed a novel global information-based link prediction algorithm which decomposes the adjacent matrix into a low rank backbone structure and a sparse noise matrix. Experimental studies indicate that this new method achieves considerable improvement when compared to each of the individual method in most of the networks
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
