Learning Cross-Domain Representation with Multi-Graph
Neural Network
Yi Ouyang
∗†
Northwestern Polytechnical
University
Bin Guo
‡
Northwestern Polytechnical
University
Xing Tang
§
Tencent
Xiuqiang He
Tencent
Jian Xiong
Tencent
Zhiwen Yu
Northwestern Polytechnical
University
ABSTRACT
Learning eective embedding has been proved to be useful in many
real-world problems, such as recommender systems, search ranking
and online advertisement. However, one of the challenges is data
sparsity in learning large-scale item embedding, as users’ historical
behavior data are usually lacking or insucient in an individual
domain. In fact, user’s behaviors from dierent domains regard-
ing the same items are usually relevant. Therefore, we can learn
complete user behaviors to alleviate the sparsity using comple-
mentary information from correlated domains. It is intuitive to
model users’ behaviors using graph, and graph neural networks
(GNNs) have recently shown the great power for representation
learning, which can be used to learn item embedding. However,
it is challenging to transfer the information across domains and
learn cross-domain representation using the existing GNNs. To
address these challenges, in this paper, we propose a novel model
-
D
eep
M
ulti-
G
raph
E
mbedding (DMGE) to learn cross-domain
representation. Specically, we rst construct a multi-graph based
on users’ behaviors from dierent domains, and then propose a
multi-graph neural network to learn cross-domain representation
in an unsupervised manner. Particularly, we present a multiple-
gradient descent optimizer for eciently training the model. We
evaluate our approach on various large-scale real-world datasets,
and the experimental results show that DMGE outperforms other
state-of-art embedding methods in various tasks.
CCS CONCEPTS
• Information systems → Recommender systems
;
• Comput-
ing metho dologies →
Neural networks; Multi-task learning;
•
Mathematics of computing → Graph algorithms.
∗
First author.
†
This work was done when Yi Ouyang worked as intern at Tencent.
‡
Corresponding author. E-mail: guobin.keio@gmail.com
§
Co-rst author.
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 the
author(s) must be honored. Abstracting with credit is permitted. 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.
Conference’17, July 2017, Washington, DC, USA
© 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ACM ISBN 978-1-4503-9999-9/18/06.. .$15.00
https://doi.org/10.1145/1122445.1122456
KEYWORDS
Cross-domain representation, item embedding, graph neural net-
work, multi-task learning, graph representation learning
ACM Reference Format:
Yi Ouyang, Bin Guo, Xing Tang, Xiuqiang He, Jian Xiong, and Zhiwen
Yu. 2019. Learning Cross-Domain Representation with Multi-Graph Neural
Network. In Proceedings of ACM Conference (Conference’17). ACM, New
York, NY, USA, 10 pages. https://doi.org/10.1145/1122445.1122456
1 INTRODUCTION
Recently, many online personalized services have utilized users’
historical behavior data to characterize user preferences, such as:
online video sites [
6
], App stores [
4
], online advertisements [
13
]
and E-commmerce sites [
32
,
37
]. Learning the representation from
user-item interactions is an essential issue in most personalized
services. Usually, low-dimensional embeddings can eectively rep-
resent attributes of items and preferences of users in a uniform
latent semantic space, which are helpful to provide personalized
services and improve user experience. Moreover, the representation
of users and items has been widely applied to many research topics
related to above real-world scenarios, including: large-scale recom-
mendation [
32
,
35
], search ranking [
5
,
11
], cold-start problem [
37
].
In large-scale personalized services, there are usually a relative
small portion of active users, and a majority of non-active users
often interact with only a small number of items, users’ behavior
data is thus lacking or insucient in an individual domain, which
makes it dicult to learn eective embeddings [
33
]. On the other
hand, though data from a single domain is sparse, users’ behav-
iors from correlated domains regarding the same items are usually
complementary [
39
]. Take the App store as an example, there are
two ways users interact (e.g., download) with items (i.e., Apps).
One is downloading Apps recommended on the homepage or cate-
gory pages of App store (i.e., recommendation domain), the other
is by searching (i.e., search domain). User behaviors in the search
domain reect user’s current needs or intention, while that in the
recommendation domain represent user’s relative long-term in-
terests. Leveraging the interaction data from the search domain
can improve the performance of recommendation. On the other
hand, interaction data from the recommendation domain can also
help to explore user’s personalized interests and therefore optimize
the ranking list in search domain. Therefore, we are motivated to
leverage the complementary information from correlated domains
to alleviate the sparsity problem.
arXiv:1905.10095v1 [cs.LG] 24 May 2019