Local, Global And Scale-Dependent Node Roles

Scholkemper, Michael

doi:10.1109/icas49788.2021.9551110

Cited by 3 publications

(1 citation statement)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This paper [40] critically re-evaluates the concept of node equivalences, such as structural and automorphic equivalence, originally devised in social network analysis and subsequently gaining significance in graph-based learning tasks. Departing from the traditional one-hop neighborhood or global graph structure definitions, the study introduces the notion of exact node roles with a scale parameter, elucidating the importance of considering the distance within which the ego network of a node influences role assignments.…”

Section: Literature Reviewmentioning

confidence: 99%

WITHDRAWN: IEER: Inferring Embedded Email Roles

Ejaz,

Cheema,

Khan

2024

Preprint

View full text Add to dashboard Cite

In this study, we investigate the use of representation learning techniques for solving complex problems represented as graphs. To this end, we propose a novel method for learning graph representations called Inferring Embedded Email Roles (IEER). Our method generates low-dimensional vectorized features by acquiring local structural information for each vertex in the graph. Unlike existing approaches that have focused on incorporating various graph attributes, our method takes a unique approach to improving the performance of the base framework, EMBedding Email-based Roles (EMBER). This research aims to understand the effectiveness of an embedding technique for generating graph representations and its suitability for replacing the base framework technique. We identify an embedding technique that significantly improves the performance of EMBER, resulting in a 94\% accuracy improvement using the Permutation Invariant technique, which has been shown to outperform matrix factorization techniques in capturing structural behaviors. This technique has been shown to outperform matrix factorization techniques in dealing with structural behaviors. Additionally, our method is scalable and linear concerning the number of vertices in the graph.

show abstract

Section: Literature Reviewmentioning

confidence: 99%