search

benedekrozemberczki / graph2vec

A parallel implementation of "graph2vec: Learning Distributed Representations of Graphs" (MLGWorkshop 2017).
https://karateclub.readthedocs.io/
GNU General Public License v3.0
891 stars 168 forks source link

Small question #2

Closed GillesVandewiele closed 5 years ago

GillesVandewiele commented 5 years ago

Hi Benedek,

First of all, massive amounts of kudos for all the implementations I discovered from you today. Strong work!

Concerning this graph2vec algorithm... At first glance, it seems identical to the rdf2vec algorithm (with the Weisfeiler-Lehman extension), except for the fact that this algorithm cannot deal with named edges (only named nodes). Am I correct, or is there a significant difference that I am currently not seeing?

Also, nice implementation of the WL relabeling!

Thanks and kind regards, Gilles

benedekrozemberczki commented 5 years ago

Yes, it is the very same idea. People also did similar implicit factorization models based on gspan and anonymous random walk patterns.

benedekrozemberczki commented 5 years ago

AWE:

https://arxiv.org/pdf/1805.11921.pdf

The paper with gspan had no public implementation.

GillesVandewiele commented 5 years ago

Ye saw that in one of your repositories as well, I'll have a closer look at it! Thanks :).

Maybe one more follow-up question:

In here, Doc2Vec is used, while in RDF2Vec, Word2Vec is used. Is there any reason for using Doc2Vec instead of Word2Vec here?