Hybrid Embedding via Cross-layer Random Walks on Multiplex Networks
ABSTARCT :
Node embedding aims to encode network nodes as a set of low-dimensional vectors while preserving certain structural properties of the network. In recent years, extensive studies have been conducted to preserve network communities, i.e., structural proximity of network nodes. However, few of them have focused on preserving the structural equivalence of network nodes, which describes the similarity of structural roles between network nodes. In this paper, we focus on a hybrid network embedding problem of how to flexibly and simultaneously preserve both structural proximity and equivalence. Here, we introduce the concept of graphlet degree vector (GDV) to describe structure roles of network nodes, and further measure structural equivalence based on their similarity. Specifically, we capture both structural proximity and equivalence by building a multiplex network, where both unsupervised and semi-supervised cross-layer random walk (CL-Walk) methods are implemented. By carrying out experiments on both synthetic and real-world datasets, we evaluate the performance of the proposed CL-Walk methods for the tasks of node clustering, node classification, and label prediction. The experimental results indicate that the CL-Walk method outperforms several state-of-the-art methods when both structural proximity and structural equivalence are relevant to specific network analytic task.
EXISTING SYSTEM :
? MultiVERSE and the other embedding methods allow learning vector representations of nodes from networks.
? We aim here to test their performance on link prediction and network reconstruction. We hence need to predict whether an edge exists between every pairs of node embeddings.
? Moreover, to our knowledge, existing approaches cannot embed multiple nodes from multiplex-heterogeneous networks, i.e. networks composed of several layers containing both different types of nodes and edges.
? However, comparisons with other approaches are not possible as, to our knowledge, no existing multiplex-heterogeneous network embedding method are currently available in the literature.
DISADVANTAGE :
? In this work, we presented Multi-Net: a basic framework on learning features in an unsupervised way from a multilayer network as an optimization problem.
? We have given a simple search strategy on the multiplex network to efficiently get a node’s context or neighborhood in a multiplex network setup.
? We construct our problem as a maximum likelihood optimization problem, similar to earlier works like word2vec , node2vec and OhmNet.
? We frame our method in such a way that it can be applied to (un)directed and (un)weighted networks also.
PROPOSED SYSTEM :
• Multi-layer networks, including multiplex and multiplex-heterogeneous networks have been proposed to handle these more complex but richer heterogeneous interaction datasets.
• We propose, to our knowledge, the first multiplex-heterogeneous network embedding method (with an embedding of the different types of nodes).
• We propose a method to evaluate multiplex-heterogeneous network embedding on link prediction. We demonstrate the efficiency of MultiVERSE on this task on two biological multiplex-heterogeneous networks.
• We propose a benchmark to compare the performance of MultiVERSE and other embedding methods for multiplex and multiplex-heterogeneous networks.
ADVANTAGE :
? We demonstrate superior performance of Multi-Net on four real-world datasets from different domains. In particular, we highlight the uniqueness of Multi-Net by leveraging it for the complex task of network reconstruction.
? We also note that Classical and Diffusive random walk methods performs better only if the network size is very small and the performance degrades when number of edges increase in the network.
? This is due to the overhead of extracting shortest paths between nodes across multiplex networks.
? Eventhough, Multi-Net and Physical random walks performs poor compared to DeepWalk, their performance improves for larger networks like Twitter.
|