Name
Abstract | ||
|---|---|---|
In the era of big data, the study of networks has received an enormous amount of attention. Of recent interest is network embedding—learning representations of the nodes of a network in a low dimensional vector space, so that the network structural information and properties are maximally preserved. In this paper, we present a review of the latest developments on this topic. We compare modern methods based on matrix factorization, including GraRep [5], HOPE [22], DeepWalk [23], and node2vec [12], in a collection of 12 real-world networks. We find that the performance of methods depends on the applications and the specific characteristics of the networks. There is no clear winner for all of the applications and in all of the networks. In particular, node2vec exhibits relatively reliable performance in the multi-label classification application, while HOPE demonstrates success in the link prediction application. Moreover, we provide suggestions on how to choose a method for practical purposes in terms of accuracy, speed, stability, and prior knowledge requirement. |
Year | Venue | Field |
|---|---|---|
2018 | DMBD | Data mining,Vector space,Computer science,Matrix decomposition,Network embedding,Big data |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
21 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Xin Liu | 1 | 3 | 1.76 |
| Kyoung-Sook Kim | 2 | 24 | 14.07 |