WalkingTime: Dynamic Graph Embedding
Using Temporal-Topological Flows
David Bayani
School of Computer Science, Carnegie Mellon University, 5000 Forbes Ave. , Pittsburgh, PA




Overview of Lightning Talk                      walk framework [2], allowing us to leverage
                                                a Skip-gram inspired back end to produce the
Node embedding - the process of generating final embeddings while simultaneously spar-
a relatively low dimensional vector to sum- ing the need to discretize or align time-related
marize a vertice’s many roles in a network attributes in the network. In our lightning
- has received increased attention in recent talk, we plan to overview WalkingTime, dis-
years, with many contemporary developments cuss its on-going evaluation across a series
building off of random-walks and NLP-inspired of tasks, and detail the potential value of both
embedding methods, specifically Skip-grams our method and the novel perspective under-
[1]. Of particular focus within the last five lying it.
years has been the development of techniques
more suitable for dynamic networks, aiming
to utilize the rich temporal structure present Acknowledgments
to better inform the embeddings produced.
Existing dynamic node embeddings, however, We would like to thank Reihaneh Rabbany
consider the problem as limited to the evolu- for her thoughts, guidance, and support while
tion of a topology over a sequence of global, conducting this work.
discrete states. Based on a fundamentally dif-
ferent handling of time, we propose a novel
embedding algorithm, WalkingTime. While
                                                References
prior works considered time as a ordered col- [1] T. Mikolov, K. Chen, G. Corrado, J. Dean,
lection of separate networks , WalkingTime          Efficient estimation of word representa-
allows for the local consideration of contin-       tions in vector space, arXiv preprint
uously occurring phenomena; while others            arXiv:1301.3781 (2013).
consider global graph snap-shots to be first- [2] A. Grover, J. Leskovec, node2vec: Scal-
order citizens , we hold flows comprised of         able feature learning for networks, in:
temporally and topologically local interactions     Proceedings of the 22nd ACM SIGKDD
as our primitives. Our temporal-topological         International Conference on Knowledge
flows eloquently extend node2vec’s random-          Discovery and Data Mining, 2016.

Proceedings of the CIKM 2020 Workshops, October 19-20,
Galway, Ireland.
" dcbayani@alumni.cmu.edu (D. Bayani)

                                                © 2020 Copyright for this paper by its authors. Use permitted under
                                                Creative Commons License Attribution 4.0 International (CC BY 4.0).
                                                CEUR Workshop                    Proceedings           (CEUR-
                                                WS.org)
1613-0073
CEURWorkshopProceedingshttp://ceur-ws.orgISSN