Entity disambiguation in anonymized graphs using graph kernels

Abstract

This paper presents a novel method for entity disambiguation in anonymized graphs using local neighborhood structure. Most existing approaches leverage node information, which might not be available in several contexts due to privacy concerns, or information about the sources of the data. We consider this problem in the supervised setting where we are provided only with a base graph and a set of nodes labelled as ambiguous or unambiguous. We characterize the similarity between two nodes based on their local neighborhood structure using graph kernels; and solve the resulting classification task using SVMs. We give empirical evidence on two real-world datasets, comparing our approach to a state-of-the-art method, highlighting the advantages of our approach. We show that using less information, our method is significantly better in terms of either speed or accuracy or both. We also present extensions of two existing graphs kernels, namely, the direct product kernel and the shortestpath kernel, with significant improvements in accuracy. For the direct product kernel, our extension also provides significant computational benefits. Moreover, we design and implement the algorithms of our method to work in a distributed fashion using the GraphLab framework, ensuring high scalability.

Name disambiguation from link data in a collaboration graph using temporal and topological features

“…We found a recent work [10] which also has a similar objective. They consider this problem in the supervised setting where they are provided with a base graph and a set of nodes labeled as ambiguous or unambiguous.…”

“…The work by Hermansson et al [10] is closely related to our work as they design a collection of graph kernels to classify multi-nodes in a supervised learning setup. Their kernels use only the graph topology, such as, graphlet counts and shortest paths, so they can be used in an anonymized network for entity disambiguation.…”

“…Besides improved performance, another advantage of our method over the methods in [10] is the superior running time. For a given node in the graph, the running time to compute the value of our features are only a few seconds, whereas computing graphlet kernel values is costly.…”

“…For a given node in the graph, the running time to compute the value of our features are only a few seconds, whereas computing graphlet kernel values is costly. In our experiments, for some of the nodes, the Matlab code provided by the authors of [10] took more than 2 days in a commodity PC. In case of unsupervised disambiguation task, our method uses two parameters that we set manually, one is the exponential decay rate (τ ) for similarity computation, and the other is α value in Equation 4.…”

“…Existing works mostly use biographical features, such as name, address, institutional affiliation, email address, and homepage; contextual features, such as coauthor/collaborator, and research keywords; and external data such as Wikipedia [7]. From methodological point of view, some of the works follow a supervised learning approach [8,10], while others use unsupervised clustering [5,9,17,25]. There exist quite a few solutions that use graphical models [3,23,26,31].…”