metapath2vec: Scalable Representation Learning for
Heterogeneous Networks
Yuxiao Dong
∗
Microso Research
Redmond, WA 98052
yuxdong@microso.com
Nitesh V. Chawla
University of Notre Dame
Notre Dame, IN 46556
nchawla@nd.edu
Ananthram Swami
Army Research Laboratory
Adelphi, MD 20783
ananthram.swami.civ@mail.mil
ABSTRACT
We study the problem of representation learning in heterogeneous
networks. Its unique challenges come from the existence of mul-
tiple types of nodes and links, which limit the feasibility of the
conventional network embedding techniques. We develop two
scalable representation learning models, namely metapath2vec and
metapath2vec++. e metapath2vec model formalizes meta-path-
based random walks to construct the heterogeneous neighborhood
of a node and then leverages a heterogeneous skip-gram model
to perform node embeddings. e metapath2vec++ model further
enables the simultaneous modeling of structural and semantic cor-
relations in heterogeneous networks. Extensive experiments show
that metapath2vec and metapath2vec++ are able to not only outper-
form state-of-the-art embedding models in various heterogeneous
network mining tasks, such as node classication, clustering, and
similarity search, but also discern the structural and semantic cor-
relations between diverse network objects.
CCS CONCEPTS
•Information systems →Social networks; •Computing method-
ologies →Unsupervised learning; Learning latent represen-
tations; Knowledge representation and reasoning;
KEYWORDS
Network Embedding; Heterogeneous Representation Learning; La-
tent Representations; Feature Learning; Heterogeneous Information
Networks
ACM Reference format:
Yuxiao Dong, Nitesh V. Chawla, and Ananthram Swami. 2017. metap-
ath2vec: Scalable Representation Learning for Heterogeneous Networks. In
Proceedings of KDD ’17, August 13-17, 2017, Halifax, NS, Canada, , 10 pages.
DOI: hp://dx.doi.org/10.1145/3097983.3098036
1 INTRODUCTION
Neural network-based learning models can represent latent embed-
dings that capture the internal relations of rich, complex data across
various modalities, such as image, audio, and language [
15
]. Social
∗
is work was done when Yuxiao was a Ph.D. student at University of Notre Dame.
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permied. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specic permission and/or a
fee. Request permissions from permissions@acm.org.
KDD ’17, August 13-17, 2017, Halifax, NS, Canada
© 2017 ACM. 978-1-4503-4887-4/17/08. .. $15.00
DOI: hp://dx.doi.org/10.1145/3097983.3098036
S. Shenker
M. I.Jordan
J. Han
A. Tomkins
R. E. Tarjan
D. Song
J. Dean
T. Kanade
R. N. Taylor
C. D. Manning
H. Ishii
H. Jensen
R. Agrawal
J. Malik
O. Mutlu
KDD
SIGGRAPH
SIGIR
FOCS
S&P
OSDI
NIPS
IJCAI
ICSE
SIGCOMM
ACL
SIGMOD
CHI
CVPR
WWW
ISCA
W. B. Croft
(a) DeepWalk / node2vec
S. Shenker
M. I.Jordan
J. Han
A. Tomkins
R. E. Tarjan
D. Song
J. Dean
T. Kanade
R. N. Taylor
C. D. Manning
H. Ishii
H. Jensen
R. Agrawal
J. Malik
O. Mutlu
KDD
SIGGRAPH
SIGIR
FOCS
S&P
OSDI
NIPS
IJCAI
ICSE
SIGCOMM
ACL
SIGMOD
CHI
CVPR
WWW
ISCA
W. B. Croft
(b) PTE
S. Shenker
M. I.Jordan
J. Han
A. Tomkins
R. E. Tarjan
D. Song
J. Dean
T. Kanade
R. N. Taylor
C. D. Manning
H. Ishii
H. Jensen
R. Agrawal
J. Malik
O. Mutlu
KDD
SIGGRAPH
SIGIR
FOCS
S&P
OSDI
NIPS
IJCAI
ICSE
SIGCOMM
ACL
SIGMOD
CHI
CVPR
WWW
ISCA
W. B. Croft
(c) metapath2vec
S. Shenker
M. I.Jordan
J. Han
A. Tomkins
R. E. Tarjan
D. Song
J. Dean
T. Kanade
R. N. Taylor
C. D. Manning
H. Ishii
H. Jensen
R. Agrawal
J. Malik
O. Mutlu
KDD
SIGGRAPH
SIGIR
FOCS
S&P
OSDI
NIPS
IJCAI
ICSE
SIGCOMM
ACL
SIGMOD
CHI
CVPR
WWW
ISCA
W. B. Croft
(d) metapath2vec++
Figure 1: 2D PCA projections of the 128D embeddings of 16
top CS conferences and corresponding high-prole authors.
and information networks are similarly rich and complex data that
encode the dynamics and types of human interactions, and are sim-
ilarly amenable to representation learning using neural networks.
In particular, by mapping the way that people choose friends and
maintain connections as a “social language,” recent advances in
natural language processing (NLP) [
3
] can be naturally applied to
network representation learning, most notably the group of NLP
models known as word2vec [
17
,
18
]. A number of recent research
publications have proposed word2vec-based network representa-
tion learning frameworks, such as DeepWalk [
22
], LINE [
30
], and
node2vec [
8
]. Instead of handcraed network feature design, these
representation learning methods enable the automatic discovery of
useful and meaningful (latent) features from the “raw networks.”
However, these work has thus far focused on representation
learning for homogeneous networks—representative of singular
type of nodes and relationships. Yet a large number of social and
information networks are heterogeneous in nature, involving diver-
sity of node types and/or relationships between nodes [
25
]. ese
heterogeneous networks present unique challenges that cannot
be handled by representation learning models that are specically
designed for homogeneous networks. Take, for example, a het-
erogeneous academic network: How do we eectively preserve
the concept of “word-context” among multiple types of nodes, e.g.,
authors, papers, venues, organizations, etc.? Can random walks,
such those used in DeepWalk and node2vec, be applied to networks
KDD’17, August 13–17, 2017, Halifax, NS, Canada