Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

Research Paper
Citation: Tsai, H.C., Chi, H.W., Shen, J.J. et coll.: Source-aware embedding training on
heterogeneous information networks. Data Intelligence 5(2023) est ce que je:
https://doi.org/10.1162/dint_a_00200
Source-Aware Embedding Training on
Heterogeneous Information Networks

Tsai Hor Chan1
1. Department of Statistics and Actuarial
Science The University of Hong Kong
Hong Kong, Chine
hchanth@connect.hku.hk

Chi Ho Wong2
2. Department of Computer Science and Engineering
The Hong Kong University of Science and Technology
Hong Kong, Chine
chwongcc@connect.ust.hk

Jiajun Shen3
3TCL Corporate Research (Hong
Kong) Hong Kong, Chine
shenjiajun90@gmail.com

Guosheng Yin1
Department of Statistics and Actuarial Science
The University of Hong Kong
Hong Kong, Chine
gyin@hku.hk

je

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

p

:
/
/

d
je
r
e
c
t
.

m

je
t
.

e
d
toi
d
n

/

je

Mots clés: Heterogeneous information network, Graph representation learning, Distribution alignment, Recommendation
système, Adversarial learning, Graph neural network

t
/

je

un
r
t
je
c
e
–
p
d

F
/

d
o

je
/

je

t

/

.

/

1
0
1
1
6
2
d
n
_
un
_
0
0
2
0
0
2
0
7
0
9
2
3
d
n
_
un
_
0
0
2
0
0
p
d

/

.

t

je

F

b
oui
g
toi
e
s
t

t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

multi-entity characteristics introduce heterogeneity to HINs and
feature different distributions among different types of entities
and relations, state-of-the-art (SOTA) methods mostly focus on
developing transformation techniques to bring feature

tasks,

to real-world

Abstrait: Heterogeneous information networks (HINs) have
been extensively applied
such as
recommendation systems, social networks, and citation networks.
While existing HIN representation
learning methods can
effectively learn the semantic and structural features in the
réseau, little awareness was given to the distribution discrepancy
of subgraphs within a single HIN. Cependant, we find that ignoring
such distribution discrepancy among subgraphs from multiple
sources would hinder the effectiveness of graph embedding
learning algorithms. This motivates us to propose SUMSHINE
(Scalable Unsupervised Multi-Source Heterogeneous Information
Network Embedding) — a scalable unsupervised framework to
align the embedding distributions among multiple sources of an
HIN. Experimental results on real-world datasets in a variety of
downstream tasks validate the performance of our method over
information network
le
embedding algorithms.

state-of-the-art heterogeneous

je. INTRODUCTION

Heterogeneous information network (HIN), also known as
heterogeneous graph, is an advanced graph data structure which
contains enriched structural and semantic
information.
Learning the representations of HINs has recently drawn
à
significant attention for
industrial applications and machine learning research.

its outstanding contribution

HINs have a variety of real-world applications including
recommendation systems [1], citation networks [2], naturel
language processing [3, 4], and social media [5, 6]. An HIN
is a multi-relation and multi-entity graph summarizing the
relations between entities, which represents a key abstraction
for organizing information in diverse domains and modelling
real-world problems in a graphical manner.

Heterogeneous information network embedding methods
aim to encode each of the entities and relations in the HIN to a
low-dimensional vector, which give feature representations to
entities and relations in the HIN. Since the multi-relation and

© 2023 Chinese Academy of Sciences. Published under a Creative Commons Attribution 4.0 International (CC PAR 4.0) Licence.

Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

distributions of different entity types and relation types to the
same embedding space [3, 7, 8].

Cependant, as of today, SOTA methods often operate on an
HIN constructed by subgraphs from multiple sources, et
most research has been based on the often implicit assumption
that the effect of distribution discrepancies among different
subgraphs on embedding learning is negligible. The major
contribution of this work is to raise awareness to the graph
learning community that this assumption does not hold in
many cases. Par exemple, graph-based recommendation sys-
tem often takes advantage of the information embedded in
HINs, where an HIN often contains a user-content interaction
graph with high-degree content entity nodes as well as a
knowledge graph with low-degree content entity nodes. Le
difference in graph structures (c'est à dire., average node degrees,
graph sizes, sparsity of connections) leads to distribution
discrepancies among subgraphs sources in the HIN. As we
will show in this paper, simply ignoring such distribution
discrepancies when training HIN embeddings would lead to
sub-optimal embedding learning performance.

Although none of the existing heterogeneous graph embed-
ding approaches attempt to solve the aforementioned problem,
there are several attempts in heterogeneous graph neural
réseaux (GNNs) that try to transfer a GNN model trained on
one graph to another [9, 10]. They often apply domain transfer
techniques to graph neural networks so that the knowledge
learned from one graph can be better transferred to another.
Note that these approaches differ from our approach in the
following important aspects: 1) Unlike the supervised learning
nature of GNN models, we are tackling the graph embedding
learning task which aims to infer node representations from
graph structures in an unsupervised manner. 2) These domain
adaption approaches often focus on adapting the learned model
of one graph to another, while we focus on how to learn one
model from a graph merged from sources.

Dans ce travail, we study the distribution discrepancy is-
sue in heterogeneous graph embedding learning. We sur-

je

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

p

:
/
/

d
je
r
e
c
t
.

m

je
t
.

e
d
toi
d
n

/

je

t
/

je

un
r
t
je
c
e
–
p
d

F
/

d
o

je
/

je

/

/

t

.

1
0
1
1
6
2
d
n
_
un
_
0
0
2
0
0
2
0
7
0
9
2
3
d
n
_
un
_
0
0
2
0
0
p
d

/

t

.

je

F

b
oui
g
toi
e
s
t

t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

learning framework

mise that simply merging sub-graphs from different sources
when training graph embeddings may negatively impact the
efficacité, which unfortunately is de facto the only known
approach to leverage data from multiple graphs. Motivated by
this limitation, we develop a scalable unsupervised multi-
source representation
learning
information network embeddings, named
heterogeneous
SUMSHINE
Unsupervised Multi-Source
(Scalable
Heterogeneous Information Network Embedding). It allows to
train
réseau
embeddings from different sources into a distribution-aligned
latent embedding space, and we confirmed that the embedding
learning performance can be significantly improved as our
framework
the distribution
discrepancy issue in learning heterogeneous information
network embeddings.

to cope with

heterogeneous

is designed

information

large-scale

pour

Our contributions can be summarized as follows:
• We study the distribution misalignment problem in HIN
embeddings and conclude that the HIN embeddings
should be trained with distribution alignment performed
on the subgraph sources of the HIN to achieve optimal
downstream task performance. To the best of our
connaissance, we are the first to introduce source-level
information
distribution alignment
network embedding.

to heterogeneous

• We propose source-aware negative sampling to balance
the training samples by source, while preserving the
scalability advantage of negative sampling. This design
overcomes the scalability constraints of existing HIN
embedding methods using GNNs.

• We validate our proposed method empirically on both link
prediction and node classification downstream tasks,
using a variety of real-world datasets. We also highlight a
practical application of our method on recommendation
systems with extensive experiments.

II. RELATED WORKS

UN. Heterogeneous Information Network Embedding

Heterogeneous information network embedding has shown
significant successes in learning the feature representations of
an HIN. Existing HIN embedding methods aim to learn a
low dimensional feature representation of an HIN. They apply
different transformation techniques to bring the embeddings
into the same latent embedding space [7, 8]. Most of the HIN
embedding methods focus on leveraging the multi-relation
characteristic in the HIN, which are known as similarity- based
méthodes [3, 4, 11, 12, 13]. Similarity-based methods are widely
adopted to learn the HIN representations by encoding the
similarity between the source and destination entities in an
bord. Within this class, there are translational methods, tel que
TransE [3], TransR [4] and TransD [11]. They take relations as
translations of the head and tail entity embeddings. Un- other
class of similarity-based HIN embedding methods uses bilinear
méthodes, such as RESCAL [14], CompleX [13], and DistMult
[12]. These methods represent relation embeddings

as a transformation of the head and tail entity embeddings [15].
There are also meta-path-based methods [16], and meta- graph-
based methods [17], utilizing the structural features in an HIN
as attempts to align the path-based or subgraph-based
distributions.

Despite their success, these works assume only one source in
the HIN and do not consider the distributional difference
among sources of subgraphs. And there is a need to align the
distributions of feature embeddings from different sources of
the HIN to improve downstream task performance. Without
loss of generality, we focus on similarity-based embedding
methods to illustrate our distribution alignment approach. Notre
method can be easily applied to all HIN embedding methods on
multi-source HINs in general as the alignment is performed on
samples of node and relation type embeddings.

Recently there are methods using GNNs to learn the
representations of an HIN [7, 9, 18, 19, 20, 21]. Although
GNNs can extract the enriched semantic information contained
dans
the HIN, the embeddings of these models are often
trained on a supervised or semi-supervised basis with respect to
a specific task. Label information on nodes and edges needs to
be provided for satisfactory embedding learning. And they can
hardly be generalized when the embeddings need to be applied
to another task. En plus, most GNN-based methods work
with the adjacency matrix of the HIN, par exemple., graph convolutional
neural network (GCN) [18] and its variants [1] on HIN perform
node aggregation based on the transformed adjacency matrix.
These matrices cannot be processed by the memory. Donc,
it is difficult to apply GNN-based HIN embedding methods for
large-scale tasks such as recommendation systems which
contain networks with billions of user nodes and millions of
movies.

In contrast,

the aforementioned similarity-based HIN
embedding methods perform embedding learning on edge
samples, which allows parallelism and therefore scalability.
Since the trained embeddings learn HIN representations by
encoding the similarity, the similarity features of the HIN are
not associated with a specific task. These properties motivate us
to propose a multi-source HIN representation
learning
framework which is not only applicable to any downstream task
but also is scalable to large HINs.

B. Distribution Alignment

Distribution alignment, also known as domain adaptation in
transfer learning, has been a key topic in HIN representation
learning, as the heterogeneity in entities and relations intro-
duces misalignments in their respective distributions. Il y a
many attempts in existing work to align the distributions of key
features in an HIN. Transformation approaches aim to learn
un
transformation matrix or an attention mechanism to
translate the feature embeddings of different types (nodes or
edges) into the same embedding space [7, 8]. Most of the
similarity-based methods mentioned above also attempt to
align the feature embeddings between entities and relations in
an HIN [3, 4, 11, 12]. Par exemple, TransE [3] approximates
distribution of the tail node embedding in an edge by the

je

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

p

:
/
/

d
je
r
e
c
t
.

m

je
t
.

e
d
toi
d
n

/

je

t
/

je

un
r
t
je
c
e
–
p
d

F
/

d
o

je
/

je

/

/

t

.

1
0
1
1
6
2
d
n
_
un
_
0
0
2
0
0
2
0
7
0
9
2
3
d
n
_
un
_
0
0
2
0
0
p
d

/

t

.

je

F

b
oui
g
toi
e
s
t

t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

sum of head and relation embeddings. Heterogeneous graph
attention network (HAN) [8] adopts a learnable transformation
layer to each node type to transform the node embeddings into
a space invariant to node types.

Adversarial learning approaches introduce discriminator
networks as a domain classifier whose losses are used to
measure high-dimensional distribution differences [10, 19, 22,
23]. De plus, several works applied distance measures such as
the maximum mean discrepancy
to perform
distribution alignment [9], these works aim to minimize the
distances between distributions to align the distributions of
feature embeddings. These alignment methods have been
extensively applied to domain adaptation to improve transfer
learning performance among multiple graphs. Cependant, ces
methods are never introduced to align the feature distributions
within an HIN.

(MMD)

Inspired by the above works in distribution alignment, nous
include both the distance measure approach and the adversarial
approach in our proposed framework. We use these alignment
methods to align the distributions of HIN embeddings with
respect to sources, in addition to their original attempts to align
the distributions of nodes or edge types. We assess the
performance of these distribution alignment methods in
aligning the embedding distributions by experiments on
different downstream tasks, such as node classification and link
prediction.

III. PROBLEM STATEMENT

UN. Definitions

Heterogeneous Information Network: A heterogeneous in-
formation network is defined by a graph 𝒢 = (𝒱, ℰ, 𝒜, ℛ) où
𝒱, ℰ, 𝒜, ℛ represent the set of entities (nodes), relations (edges),
entity types, and relation types, respectivement. A triple in E is
defined by e = (h, r, t), where h, t ∈ 𝒱 are the heads and tails
nodes representing the entities in 𝒢, and 𝑟 ∈ ℛ represents the
type of relation connecting the entities. For 𝑣 ∈ 𝒱, 𝑣 is mapped
to an entity type by a function 𝜏(𝑣) ∈ 𝒜, and 𝑟 is mapped to a
relation type by a function 𝜙(𝑟) ∈ ℛ.

Heterogeneous Information Network Embeddings: Nous
encode the similarity of each node in the HIN to a 𝑑-
dimensional vector with a similarity function 𝑓(𝑒). The node
and edge type embeddings can be used as input features for
training an arbitrary downstream task model.

IV. Méthodologie

We introduce SUMSHINE in this section. The major com-
ponent of SUMSHINE consists of a source-aware negative
sampling strategy and a loss function designed to regularize
distribution discrepancies across subgraphs. Conceptual
visualization of the training paradigm of SUMSHINE is shown
in Figure 1.

UN. Source-Aware Negative Sampling

Given a positive edge e = (h, r, t), negative sampling re-
places either a head or a tail (but not both) by another arbitrary
node in the HIN to produce negative edges which do not exist
in the original graph [3, 5]. The embeddings can be learned by
maximizing the distance between the positive samples (c'est à dire.,
ground truth edges) and the negative samples. Cependant,
sampling from imbalanced subgraphs leads to data imbalance
problem between subgraph sources. Edges in larger subgraphs
(such as a user interaction graph) are sampled more often than
the smaller subgraphs (such as an album knowledge graph). À
rebalance the data with respect to sources, we introduce source-
aware negative sampling to sample edges uniformly from each
subgraph source. By source-aware sampling we can balance the
number of edges sampled by sources, and reduce the bias on
embeddings from data imbalance. For each subgraph source 𝒢7
in 𝒢, we sample a fixed-size batch of edges from it to match the
dimensions of sample embedding matrices. Given and edge
𝑒7 = (ℎ7, 𝑟7, 𝑡7) from a source 𝒢7 , we select a set of negative
G or a tail node
samples 𝑆EF
G within the
G or 𝑡7
by 𝑡7
subgraph. We denote the set of negative samples as

G by replacing either a head node by ℎ7

G are entities other than ℎ7

G , where ℎ7

G and 𝑡7

G = {(ℎ7
𝑆EF

G, 𝑟7, 𝑡7)| ℎ7

G ∈ 𝒱>} ∪ {(ℎ7, 𝑟7, 𝑡7

G)| 𝑡7

G ∈ 𝒱>}.

The negative samples are combined with a batch of positive
edges to compute the similarity on a mini-batch basis. Le
similarity-based loss function is given by

;

ℒJ7K = L L L [𝑓(𝑒7) − 𝑓(𝑒7
S∈TUF
EF

EF∈𝒢F

79:

S

G) + 𝛾]R.

,

where 𝛾 is the margin and [𝑥]R = max (X, 0). The scoring
function 𝑓(𝑒) is uniquely defined by the HINE method. Nous
assume the embeddings of the edge samples are independent
and identically distributed (IID). We use mini-batch gradient
descent [5] to back-propagate the similarity loss to the
embeddings to learn the HIN representation.

B. Problem: Multi-Source Heterogeneous Information Net-

B. Aligning Sources with Regularization

work Embeddings

Consider a heterogeneous information network 𝒢 =
(𝒱, ℰ, 𝒜, ℛ), let 𝒮 represents the set of sources in 𝒢. We have
;
; = {(𝒱=, ℰ=, 𝒜=, ℛ>)}79:
a series of 𝐾 = |𝒮| subgraphs {𝒢7}79:
as the predefined sources of 𝒢. Let 𝒳 be the embeddings
space of nodes and edge types in 𝒢, and let 𝒳7denote the
embedding space of nodes and edge types in each subgraph
𝒢7.

We wish to assign an embedding x ∈ X to each node and edge
;
type in 𝒢. We also wish to align the distributions of {𝒳7}79:
such that for a model ℳ trained on graph 𝒢, on a given
downstream task 𝒯, the model ℳ can perform accurately.

As mentioned above, one of the key issues we want to ad- dress
here is to alleviate the distribution discrepancies among different
subgraphs. More specifically, given an arbitrary pair of
; , we define the distribution functions 𝑃 and
subgraphs in {𝒢7}79:
𝑄 on the embedding space to be the embedding distributions on
the two subgraphs, and we aim to encourage less distribution
discrepancy between 𝑃 and 𝑄 despite their domain differences.
To achieve this, we introduce two regularization methods —
adversarial
distance-measure-based
regularization.

regularization

et

je

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

p

:
/
/

d
je
r
e
c
t
.

m

je
t
.

e
d
toi
d
n

/

je

t
/

je

un
r
t
je
c
e
–
p
d

F
/

d
o

je
/

je

/

t

/

.

1
0
1
1
6
2
d
n
_
un
_
0
0
2
0
0
2
0
7
0
9
2
3
d
n
_
un
_
0
0
2
0
0
p
d

/

t

.

je

F

b
oui
g
toi
e
s
t

t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

figue. 1: Training paradigm of our proposed SUMSHINE HIN Embedding method.

We first introduce distance-measure-based regularization. Dans
this paper, we adopt the distance measures MMD [24], le
Kullback–Leiber (KL) divergence, and the Jensen–Shannon
(JS) divergence [25] in our experiments, while our framework
can be generalized to incorporate any distance measures. Nous
use ∆ to denote the distribution distance between P and Q.
The KL divergence on P and Q is defined as

Δ;Z(𝑃 ∥ 𝑄) = L 𝑃(𝑥) log _

a∈𝒳

𝑃(𝑥)
𝑄(𝑥)

,
‘

and the JS divergence is the symmetric and smoothed version
of the KL divergence defined by

ΔbT(𝑃 ∥ 𝑄) =

1
2

eΔ;Z(𝑃 ∥ 𝑄) + Δ;Z(𝑄 ∥ 𝑃)F.

The MMD loss is a widely used approach to alleviate the
marginal distribution disparity [26]. Given a reproducing kernel
Hilbert space (RKHS)
[24], MMD is a distance measure
between P and Q which is defined as

𝑀𝑀𝐷(𝑃 ∥ 𝑄) = i𝜇k − 𝜇li

ℋ

,

where 𝜇k and 𝜇lare respectively the kernel means computed
on 𝑃 and 𝑄 by a kernel function 𝑘(∙) (par exemple., a Gaussian kernel).
We perform distribution alignment between pairs of subgraphs.
For each batch sampled by source-aware sampling and each
pair of sources, we compute the distribution differences of
embeddings for both relation types and entities, using one of
the distance measures introduced above. The regularization loss
ℒp7Jq is the sum of distribution distances for both entity and
relation type embeddings over each pair of sources. The total
loss can be obtained by combining ℒp7Jq with the similarity loss
to propagate both the similarity and the distribution discrepancy
into HIN embedding training,

ℒqrq = ℒp7Jq + 𝜆ℒJ7K (1)

where λ is a tuning parameter.

Alignment methods based on distance measures heavily
relies on the measure chosen, and the high dimensional
distribution difference such as the geodesic difference may not
be incorporated by the measure. R Connor et. al [27] suggested
that the high dimensionality of data in the metric space may
cause metrics of distribution differences
to be biased.
Adversarial Regularization, on the contrary, uses a feedforward
network as a discriminative classifier
le
distributional differences in high dimension to avoid comparing
high-dimensional data
the metric space directly and
ameliorate the bias compared to the aforementioned distance
measures [28].

to capture

dans

With the recent development of adversarial distribution
alignment [10, 23, 28, 29], nous
introduce adversarial
regularization to HIN embedding training. We consider the
embed- dings from different subgraphs trained by an HIN
embedding method as the generated information, and use an
adversarial discriminator D as a domain classifier to classify the
source of the embeddings. Par conséquent, we consider the loss from
the discriminator a measure of distribution discrepancy
between the sources and use it to align the embeddings
distributions from different sources [10, 28].
Let ℬ7 ⊆ 𝒳7be the node and edge type embeddings in a sampled
batch from a subgraph source 𝒢7. The discriminator receives the
batch of embeddings ℬ7 , and generates the probability that
which source these embeddings are from. The predictions are
compared with the ground truth one-hot label 𝑦7, where its i-th
entry is 1 with the rest being zeros. The loss of the discriminator
is given by

ℒw = L 𝔼yF∈ℬF
>∈𝒮

[𝐷(𝑥7) − 𝑦7].

We then compute the adversarial loss and combine it together
with the similarity loss. We compute the distribution distance
by inverting the true label 𝑦7 to 𝑦z where 𝑖 ≠ 𝑗. The adversarial
loss is then given by

je

D
o
w
n
o
un
d
e
d

F
r
o
m
h

t
t

p

:
/
/

d
je
r
e
c
t
.

m

je
t
.

e
d
toi
d
n

/

je

t
/

je

un
r
t
je
c
e
–
p
d

F
/

d
o

je
/

je

/

t

/

.

1
0
1
1
6
2
d
n
_
un
_
0
0
2
0
0
2
0
7
0
9
2
3
d
n
_
un
_
0
0
2
0
0
p
d

/

t

.

je

F

b
oui
g
toi
e
s
t

t

o
n
0
7
S
e
p
e
m
b
e
r
2
0
2
3

Data Intelligence Just Accepted MS.
https://doi.org/10.1162/dint_a_00200

ℒ~p(cid:127) = L 𝔼yF∈ℬF
>∈𝒮

(cid:128)𝐷(𝑥7) − 𝑦z(cid:129).

given node from source 𝑖 with embedding 𝑧 in the shared-
semantic embedding space 𝑍 and ℳ7 be the downstream
task model trained to make prediction 𝑦(cid:156) = ℳ7(𝑧) [10],

The loss 𝐷(𝑥7) − 𝑦z for each pair of sources 𝑖 and 𝑗 indicates
the distributional difference between them. We aim to include
this adversarial loss to the embeddings such that the
embeddings can be more similar in distribution to fool the
discriminator. We then multiply the adversarial loss by the
tuning parameter 𝜆 and compute the aggregated loss using
equation (1) with ℒp7Jq replaced by ℒ~p(cid:127).

ℒ7 = (cid:157) 𝑝7(𝑧)Δ(𝑃(cid:151)7(𝑦|𝑧), 𝑃(𝑦|𝑧))𝑑𝑧

,

(cid:158)

where Δ is the divergence function to determine the loss of
predicted labels to ground truth labels. We have the following
theorem:

Theorem 1 If the following conditions are satisfied:

V. THEORETICAL ANALYSIS

(cid:135)𝑃(cid:151)7(𝑦|𝑧) − 𝑃(cid:151)z(𝑦|𝑧)(cid:135) < 𝑐7z, ∀𝑖, 𝑗 ∈ 𝒮 We provide theoretical analysis to show why aligning the distribution of embeddings from different sources of sub- graphs in a heterogeneous graph can improve the downstream task performance, where the error of generalization will be bounded by probability with an optimized bound. Settings We first define the loss of generalization. When generalizing the model from a origin environment to the target environment on the same task 𝒯 , we want to minimize the generalization bound 𝜖 such that the error of generalization is bounded by 𝜖 in probability, which is for any 𝛿 > 0,

ℙe(cid:135)ℒr(cid:136)(cid:137) − ℒpEJq(cid:135) ≤ 𝜖f ≤ 1 − 𝛿,

where ℒr(cid:136)(cid:137), ℒpEJq are the losses of the origin and destination
respectively for the downstream task 𝒯and (cid:135)ℒr(cid:136)(cid:137) − ℒpEJq(cid:135) est
the error of transferal. We further assume that the source
discrepancy leads to the largest generalization error than any
pairs of subgraphs in 𝒢, which is formulated in assumption 1.

Assumption 1 Suppose {𝒢}79:
subgraph sources of 𝒢, let 𝒢J(cid:139)
{𝒢}79:

; which has the largest generalization error,

; is the set of 𝐾 pre-defined
∗, 𝒢J(cid:141)

∗be the pair of subgraphs in

(cid:142)ℒ𝒢(cid:143)(cid:139)

∗ − ℒ𝒢(cid:143)(cid:141)

∗ (cid:142) ≥ (cid:142)ℒ𝒢(cid:145)(cid:139) − ℒ𝒢(cid:145)(cid:141)

(cid:142) ∀𝑠:, 𝑠(cid:148) ∈ 𝒮,

(cid:135)𝑝7(𝑧) − 𝑝z(𝑧)(cid:135)
𝑝7(𝑧)

< 𝜖7z, ∀𝑖, 𝑗 ∈ 𝒮 Then we have ; ; L L(ℒ7 − ℒz) 79: 7¡z z9: ; ; ≤ L L(ℒ7𝜖7z − 𝑐7z + 𝑐7z + 𝜖7z) . 79: 7¡z z9: Theorem 1 states that if we want to control the generalization loss from each source i to any other sources, we need to align both the semantic meaning and the distributions 𝑝7(𝑧) of the embeddings by controlling every pairwise distance. The proof of theorem 1 is given by Zhang et al. [10]. Since all the subgraphs are trained jointly and the subgraph embeddings are essentially having the same semantic meaning, we further assume 𝑐7z → 0 in theorem 1 as the embeddings having very close semantic meanings (i.e., ground truth labels will be the same for a given 𝑧). Then we have the following corollary: Corollary 1 If 𝑐7z → 0, we have the following reduced version of theorem 1: ; ; L L(ℒ7 − ℒz) 79: 7¡z z9: ; ; ≤ L L ℒ7𝜖7z . (2) 79: 7¡z z9: ∗ − ℒ𝒢(cid:143)(cid:141) ∗ (cid:142). This assumption is where ℒ(cid:137) is the downstream task loss using a graph 𝑔, then we assume that for any pair of subgraph in 𝒢, the generalization loss is less than or equal to (cid:142)ℒ𝒢(cid:143)(cid:139) reasonable since the sources of subgraphs are mostly having the largest semantic difference and least overlaps. We can focus on minimizing the source-level embedding distribution discrepancy with this assumption. To obtain a theoretical bound of source-level generalization error, we generalize the current pairwise analysis from Zhang et al. [10] to multiple sources. Given a specific downstream 𝒯 and a series of true labels ; (cid:150)𝑃(cid:151)7(𝑦|𝑧)(cid:153) task loss for source 𝑖, and 𝑝7(𝑧) is the density function of a for each source 𝑖 in 𝒮. ℒ7 is the downstream 79: Equation (2) indicates that in order to reduce the generalization error between any pairwise environments, we only need to minimize the distribution difference of all pairs of environments. In other words, we want to minimize which can be achieved by minimizing the adversarial loss ℒ~p(cid:127). On the other hand, the similarity loss ℒJ7K can highlight the node and edge features in the graph, thus ℒ7 can still be minimized. VI. THEORETICAL ANALYSIS A. Datasets We collect public datasets for benchmarking HIN embedding methods that contain multiple sources: WordNet18 (WN18) [3] and DBPedia (DBP). Table II provides a summary of the datasets used for experiments. We also compose a real l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i / t . / 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d . t / i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 TABLE I: Example heterogeneous graph embedding methods and their scoring functions Model Embedding Space Relation Embeddings Scoring Function Space Complexity TransE [3] TransR [4] TransD [11] RESCAL [14] DistMult [12] ComplEx [13] h, t ∈ Rd h, t ∈ Rd h, t, Mh, Mt ∈ Rd h, t ∈ Rd h, t ∈ Rd h, t ∈ Cd r ∈ Rd r ∈ Rd, Mrt ∈ Rk×d r, Mr ∈ Rd Mrt ∈ Rd×d r ∈ Rd r ∈ Cd −∥ e𝑴(cid:136)𝑴𝒉 𝑻 + 𝑰f𝒕 ∥“ ∥h + r – t∥𝒑 ∥ 𝑴𝒓𝒉 + 𝒓− 𝑴𝒓𝒕 ∥𝒑 𝑻 + 𝑰f𝒉 + 𝒓 − e𝑴(cid:136)𝑴𝒉 𝒉𝑻𝑴𝒓𝒕 𝒉« 𝒅𝒊𝒂𝒈(𝒓)𝒕 𝒉« Ree𝒅𝒊𝒂𝒈(𝒓)f𝒕 O(d) O(d2) O(d2) O(d2) O(d) O(d) h, r, t: embeddings of head, relation, and tail; d: dimension of the embedding vector; Re(z): real part of complex number z; diag(x): diagonal entries of matrix x; Cd: complex space of dimension d; Mr, Mt: learnable matrices to transform the relation or tail embeddings. TABLE II: Datasets Summary A. Benchmarking Methods Dataset DBP-Total1 DBP-WD2 DBP-YG3 MRec-Total MRec-Album MRec-User WN18-Total WN18-A WN18-B WN18-C |𝒱| 118,907 42,201 37,805 284,908 57,203 235,693 40,943 39,398 20,179 7,516 |ℛ| 305 259 236 8 5 3 18 6 7 5 |ℰ| 118,907 60,000 60,000 307,029 62,915 246,629 151,442 96,598 41,836 13,008 |𝒜| 1 1 1 1 1 1 1 1 1 1 1 Total: The whole graph constructed by merging the sub- graph sources 2 WD: Wikidata source of DBPedia 3 YG: WordNet source of DBPedia dataset, namely MRec (Movie Recommendation), based on real user movie watching data from a practical recommendation system. MRec has two sources — one is representing the user-movie interaction graph, containing the users’ movie- watch histories, and another one is simulating the knowledge graph of the album of movies with ground truth entities related to the movies such as tags, directors, and actors. We use the MRec dataset to model the distribution difference caused by graph sizes in HINs. To validate the performance when our method is applied to more than two sources, we perform experiments on the WN18 dataset which contains three sources of subgraphs — namely A, B, and C. The subgraphs are created by categorizing the relations according to their semantic meanings so that different subgraphs will correspond to different sets of relations, incurring different average node degree per relation type. Details on the sources can be found in the Appendix. For node classification, we collect channel labels from the MRec dataset for 7000 movie nodes where these nodes are present in both the user interaction graph and album knowledge graph. Each movie node is labelled by one of the following six classes: “not movie”, channel 1 to 4, or “other movie” (i.e. channel information not available). We additionally sample 3000 “not movie” (i.e., negative) entities from the MRec data for training in order to produce class-wise balanced data. We randomly choose 7000 movie entities and 3000 non-movie entities from the testing graph as the testing data. We compare our method against the baseline HIN embed- ding learning methods, including TransE [3], TransR [4], and DistMult [12], and validate the improvements provided by our method. We also show the performance of GNN-based approaches [18, 20, 21], of which the main goal is to learn node embeddings for a specific downstream task, as a reference. For simplicity, we use the scoring function of TransE [3] in our proposed framework, while the performance of our method with other scoring functions is presented by ablation studies in Section VII-A. To validate the effectiveness of our approach, we apply the node and edge type embeddings produced by each approach as the feature input to downstream tasks. Table I presents a summary of the embedding methods and their scoring functions Descriptions of each method are listed below: • TransE [3]: Learning the relations in a multi-relation graph by translating the source and destination node embeddings of the relation. • TransD [11]: In addition to TransR translating the relation space, TransD also maps the entity space to a common latent space. • TransR [4]: Building entities and relations in separate embedding spaces, and project entities to relation space then building translation between the projected entities. • RESCAL [14]: RESCAL is a bilinear model that captures latent semantics of a knowledge graph through associate entities with vectors and represents each relation as a matrix that models pairwise interaction between entities. Entities and relations are represented as a multi- dimensional tensor to factorize the feature vectors to rank r. • DistMult [12]: Improving the time complexity of RESCAL to linear time by restricting the relationship to only symmetric relations. B. Experiment Settings We perform inductive link prediction [30] as the down- stream task to validate our framework. After we obtain the node and edge type embeddings produced by different HIN embedding approaches, we use a multiple layer perceptron (MLP) matcher model to perform the downstream task. A matcher model is a binary classifier that output the probability of having a link given the edge embedding (i.e., concatenated l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i . t / / 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d t / . i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 embedding of head, tail and relation) as the input. For GNN baselines, we directly train a GNN to perform link prediction instead of MLP. A matcher model can perform inductive link prediction across subgraphs rather than transductive [30] link prediction which can only predict linkage with the observed data (i.e., the subgraph used for training). To highlight the advantage of combining subgraphs and learning embeddings in distribution-aligned latent embedding space, we design an experiment setting for inductive link prediction as follows: When training for the downstream tasks, we only take the training data that contain edges from one subgraph while keeping the data which contain edges from other graphs as evaluation data. Note that we borrow this setting from the literature on GNN transfer learning [9, 10], where the goal of these works is to transfer the GNN model from one graph to another. However in our setting, rather than showing how transferrable the downstream task models are, we show how a distribution-aligned embedding training mechanism can benefit the downstream task performance, especially when subgraphs. For there are distribution demonstration of results, we denote the training-testing split in each link prediction experiment with an arrow “Training → Testing” for notational convenience. shifts among For each testing edge, we replace the head and then the tail to each of the 1000 negative entities sampled from the testing entities. We rank the true edge together with its negative samples according to the probability that an edge exists be- tween the head and tail output from the MLP matcher model. We sample 1000 negative entities to corrupt the ground truth edge instead of all the testing entities in the subgraph because scaling the metrics can enhance the comparability among datasets. Since each testing entity has an equal probability to be replaced, the downstream task performance is not affected by the choice of the sample size of negatives. We use node classification as another downstream task. We first train an MLP node classification model on one source of subgraph and then test the model on another source. The classification model takes an HIN node embedding as the input, and classify the node to one of the six classes according to its embedding. We evaluate the link prediction performance using Hits@n and mean reciprocal rank (MRR) and the node classification performance using classification accuracy. More details on evaluation metrics and model configurations are presented in the appendix. C. Link Prediction We validate our framework by inductive link prediction. Table III provides a summary of the prediction performance of our method to various baselines. We choose the JS divergence to be the distance measure for alignment. More discussions on the effects of different distance measures will be included in section VII-B. The experiments are performed on MRec and DBPedia datasets with two sources. We observe that the link prediction results after distribution alignment, with either or regularization adversarial distance-measure-based regularization, are uniformly better among the benchmarks for all evaluation metrics. The performance of adversarial regularization is superior to the JS divergence, which supports the superiority of adversarial alignment over distance- measured-based alignment. The results show that inductive link prediction is optimized for multi-source graphs if we align the distributions of the embeddings. We also observe that GNN models underperform our method in most of the inductive link prediction tasks. GNN link prediction models can extract global features by aggregating node features from the whole graph (e.g., through the transformed adjacency matrix), which is more capable than similarity-based method focusing on local similarity features. However, the misalignment in subgraph sources still decrease the performance of GNN-based link prediction models, which make them underperform our model in general. Additionally, out-of-memory errors were reported for GNN models when the size of user graph is doubled in the User→Album experiment. This highlights the scalability constraints of GNN models. We further validate our framework on datasets with more than two sources. Table IV presents the results of inductive link prediction performance for each of the six training-testing splits on the WN18 dataset. We observe that in most of the tasks the performances are improved with distribution aligned embeddings. This validate the consistency of our framework when K is generalized to be larger than 2 (i.e., multiple sources). The MRec dataset is simulated to have a significant imbalance of data with respect to sources. Hence the data without source-aware sampling are mostly sampled from the user interaction graph and only a few of them are from the album knowledge graph. It is noteworthy that since the user- interaction graph is sparse (i.e. as users have diverged interests), the link prediction model trained on the album knowledge graph, is heavily biased and less transferrable to the user-interaction graph, leading to a performance which occasionally worse than a random guess. With source-aware sampling, smaller subgraphs can be sampled equal times to larger subgraphs. Therefore, the information in the smaller subgraphs can be leveraged especially when there is a large degree of data imbalance among the sub- graphs. Hence source-aware sampling significantly increases the awareness to small subgraphs, which resolves the data imbalance problem in existing methods. D. Node Classification Table V presents the node classification performance with or without distribution alignment respectively. We observe that there are improvements in accuracy for both user to album and album to user transferal tasks. Note that the MRec dataset contains subgraphs with significant different average node degrees. Therefore, without taking into account the imbalance issue, the node and edge type embeddings will be dominated by the semantic information contained in the user interaction graph. With the help of distribution alignment l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i / / t . 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d / t . i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 TABLE III: Link prediction performance of SUMSHINE to baseline methods on DBPedia and MRec datasets (JS: Regularization loss is the JS Divergence; ADV: Regularization loss is the adversarial loss). DBP MRec Model TransE [3] TransR [4] GIN [20] GCN [18] GAT [21] SUMSHINE-JS SUMSHINE-ADV YG→WD User→Album WD→YG Album→User MRR ↑ Hit@10↑ MRR Hit@10 MRR Hit@10 MRR Hit@10 0.0067 0.0130 0.0044 0.0295 0.1290 0.0293 0.0981 0.0276 0.0674 0.0368 0.1040 0.0474 0.3386 0.0487 0.0055 0.1100 0.0559 0.1643 0.0834 0.0262 0.1549 0.0060 0.0670 0.0275 0.1337 0.0414 0.0149 0.1232 0.0232 0.0632 0.0498 0.0435 0.0553 0.0653 0.1022 0.0117 0.0302 0.0259 0.0244 0.0288 0.0320 0.0536 0.0232 0.0638 0.0607 0.0511 0.0653 0.1236 0.1257 0.0059 0.0051 0.0587 0.0608 0.0305 0.0380 0.1954 TABLE IV: Link prediction performance of our method to TransE on the WordNet18 dataset which has three sources. The similarity loss used for SUMSHINE is the same as TransE. TransE SUMSHINE Data Split MRR↑ Hit@10↑ MRR Hit@10 0.0089 0.0079 0.0103 0.0100 0.0092 0.0072 0.0080 0.0087 0.0075 0.0099 0.0073 0.0069 0.0064 0.0061 0.0067 0.0104 0.0126 0.0115 0.0136 0.0138 A → B B → A B → C C → B A → C C → A 0.0081 0.0134 0.0086 0.0123 during embedding training, the structure information in the movie-knowledge graph can be leveraged and ameliorate the domination of the user-interaction graph, hence the recall may be higher while the precision is sacrificed to adjust the bias caused by a large difference in average node degrees. respectively, which is shown in Table VI. We observe that the is decreased significantly after distribution discrepancy adversarial alignment. According to the flat-minima hypothesis [31], smooth regions are the key for smooth transferal of the features between distributions, which allow better alignments in features from the subgraphs. The downstream task models can hence make use of the aligned features to improve their performances. TABLE VI: JS Divergences of the trained embeddings of DBP and MRec with respect to their sources (User and Album for MRec; WD and YG for DBP). The comparison is performed between distribution-aligned embeddings (SUMSHINE) and the original embeddings (TransE). Data Split TransE SUMSHINE User — Album WD — YG 8.7493 0.5870 0.0831 0.1522 TABLE V: Node classification performance (in classification accuracy) of TransE with or without distribution alignment respectively. — ABLATION STUDIES Data Split TransE SUMSHINE User → Album Album → User 0.5392 0.5497 0.5548 0.6249 E. Visualization To validate the performance of our alignment method, we use Isomap plots to visualize the trained embeddings with or without distribution alignment, respectively. The high- dimensional information such as geodesic distance can be preserved by Isomap when reducing the dimension of the embedding distribution. Figure 2 shows the Isomap plot of the embedding trained by TransE and SUMSHINE on the DBPedia dataset and the MRec dataset. More visualizations are shown in the appendix. It is observed that with distribution alignment, the distributions of embeddings in YG and WD are smoother (i.e. having fewer random clusters and more flat regions), while the source-invariant features such as modes of distributions are still preserved by similarity learning. The alignment in distributions can also be validated quantitatively by computing the JS divergences without and with adversarial regularization A. Impact of Scoring Functions We experiment with other HIN embedding methods by exploring different scoring functions. Table VII demonstrates the performance on link prediction using the embeddings with or without distribution alignment respectively. Similar to that of TransE, we observe that distribution alignment can still improve inference performance when the scoring function is altered. We can verify that the performance of our framework is invariant to the changes in scoring functions, which indicates that by training distribution-aligned HIN embeddings the downstream tasks can perform more accurately with any chosen scoring function. This ensures the extensibility of our framework when new HIN embedding methods are developed. B. Impact of Distance Measures We further evaluate the performance of our model when the distance measure is changed to another one, e.g., the KL divergence or MMD. Table VIII presents the performance of our framework on link prediction on the DBPedia dataset when using different distance measures. We observe that both distance measures can align the distributions of embeddings and improve the downstream task performance. Since MMD l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i . / t / 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d . t / i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 (a) WD No DA (b) YG No DA (c) WD With DA (d) YG With DA (e) User No DA (f) Album No DA (g) User With DA (h) Album With DA Fig. 2: Isomap plots of the embeddings of DBP by sources WD and YG, and MRec by sources User and Album, with or without distribution alignment (DA) respectively. The alignment method used is adversarial regularization. TABLE VII: Link prediction performances of different similarity functions. The alignment method is adversarial regularization. TABLE VIII: Link prediction performances of alignment using different distance measures on the DBPedia dataset. The similarity loss for SUMSHINE the same as TransE. Model User → Album Hit@3 MRR Album → User MRR Hit@3 Model WD → YG YG WD → MRR Hit@3 MRR Hit@3 TransE SUMSHINE-TransE TransR SUMSHINE-TransR TransD SUMSHINE-TransD DistMult SUMSHINE-DistMult RESCAL SUMSHINE-RESCAL 0.0064 0.1232 0.0670 0.1904 0.0059 0.1381 0.0206 0.0238 0.0053 0.0428 0.0024 0.1421 0.0758 0.2297 0.0023 0.1703 0.0169 0.0245 0.0022 0.0495 0.0084 0.1954 0.0051 0.2208 0.0088 0.1144 0.0448 0.0688 0.0326 0.1109 0.0030 0.2358 0.0020 0.2850 0.0047 0.1567 0.0555 0.0704 0.0350 0.1300 features computes the distribution distance in a Hilbert space [24], it can incorporate higher-dimensional the KL divergence, the link prediction performance with MMD is better than that with the KL divergence, while the time complexity of MMD is higher. We conclude that using distance measures can align the embedding quality, with small variations to the distance measure selected. the distributions and improve than C. Impact of Subgraph Sizes The difference in sizes among the subgraphs highlights the distribution discrepancies. The aforementioned size difference between the user-interaction graph and album knowledge graph is a typical example of the size difference. We further study how SUMSHINE performs as the ratio of sizes (in the number of edges) changes. We compose different variants of the MRec dataset with different ratios of the total number of TransE SUMSHINE-KL SUMSHINE-MMD SUMSHINE-JS 0.0130 0.0283 0.0471 0.0474 0.0117 0.0221 0.0534 0.0518 0.0064 0.0251 0.0280 0.0320 0.0056 0.0235 0.0204 0.0330 edges — from an approximately equal number of edges to large differences in the total number of edges. We compare link prediction performance with the original TransE and the distribution-aligned version with adversarial regularization and use MRR as the evaluation metric. Figure 3 demonstrates a decreasing trend of MRR as the ratio (album: user) of the number of edges changes from 1:4 to 1:1, which indicates that our framework has better performance when the size between subgraphs has larger differences. On the other hand, TransE is having improving performance as the number of edges of the subgraphs are close to each other. However, the link prediction performance of TransE is still lower without distribution alignment. The reason is that the user interaction graph has diverged features where the features cannot be smoothly transferred without distribution alignment. For application on graph-based recommendation systems where the user interaction graph and album graph typically have a large difference in graph size, our framework performs better to resolve the information misalignment problem for better recommendation performance. This is a practical l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i / t . / 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d t / . i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 — CONCLUSION AND FUTURE WORK We propose SUMSHINE — a scalable unsupervised multi- source graph embedding framework on HINs, which is shown to improve the downstream task performance on the HIN. Extensive experiments have been performed on real datasets and different downstream tasks. Our results demonstrate that the embedding distributions in the subgraph sources of the HIN can be successfully aligned by our method. We also show by ablation studies that the our framework is robust when the distance measure or is altered. the scoring Additionally, we show that our framework performs better when the sources are having a larger difference in the graph size. function Our framework can be further generalized to integrate multimodal HIN embeddings by aligning the distributions of the side information embeddings such as image or text embeddings. Incorporating multimodality opens the possibility of practical application of our framework to common-sense knowledge graphs where the graph is constructed by merging numerous knowledge bases, including text and image features. ACKNOWLEDGMENT We thank the anonymous reviewers and Dr. Pan Yi Teng for their insights and advice on this research. This work was partially supported by the Research Grants Council of Hong Kong (17308321) and the HKU-TCL Joint Research Center for Artificial Intelligence sponsored by TCL Corporate Research (Hong Kong). REFERENCES [1] Kojima, R., Ishida,S. Ohta, M. et al.: KGCN: A graph- based deep learning framework for chemical structures. Journal of Cheminformatics 12, 1–10 (2020). [2] Hu,W., Fey, M., Zitnik, M. et al.: Open graph benchmark: Datasets for machine learning on graphs. arXiv preprint arXiv:2005.00687 (2020). [3] Bordes, A., Usunier, N., Garcia-Duran, A. et al.: Translating embeddings for modeling multi-relational data. Advances in Neural Information Processing Systems 26, (2013). [4] Lin, Y., Liu, Z., Sun, M. et al.: Learning entity and relation embeddings for knowledge graph completion. In: Twenty- ninth AAAI Conference on Arti- Ficial Intelligence, (2015). [5] Lerer, A., Wu, L., Shen, J. et al.: PyTorch-BigGraph: A large-scale graph embedding system. In: Proceedings of the 2nd SysML Conference, (2019). [6] Gottschalk, S., Demidova, E. A multilin- gual event- centric temporal knowledge graph. In: European Semantic Web Conference. pp. 272–287 (2018). [7] Hu, Z., Dong, Y., Wang, K. et al.: Heterogeneous graph transformer. In: Proceedings of The Web Conference, pp. 2704–2710 (2020). Fig. 3: Performance (in MRR) of album→user link prediction task of SUMSHINE-ADV with respect to different ratios of the number of edges (A:U) between the album knowledge graph (A) and the user interaction graph (U). Fig. 4: Link prediction performance (in MRR) of SUMSHINE- ADV on DBPedia dataset with respect to different values of λ. Here the YG source is the training set and the WD source is the testing set. insight of the above results on the industrial application of our framework. D. Impact of Tuning Parameter λ We study the impact of tuning parameter λ in equation (1) on the performance of our method. We explore a grid of values of λ: [0.01, 0.1, 1, 10, 100, 1000] and perform adversarial distribution alignment with each λ value. Figure 4 shows how our methods perform on DBPedia datasets with different values of λ. We observe that the optimal performance is obtained when λ is 1. We also observe that the link prediction performance is worse when λ is too small or too large. When λ is too large, the regularization on the embeddings is too heavy such that the similarity feature is not preserved by the embeddings, and the lack of similarity features will decrease the link prediction performance. On the other hand, when λ is too small, the misalignment in distribution is not penalized by the alignment loss, the distribution misalignment will also decrease the link prediction performance. Hence λ should carefully be tuned to achieve optimal downstream task performance. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i / / . t 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d t / . i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 [8] Wang, X., Ji, H., Shi, C.B. et al.: Heterogeneous graph attention network. I n: The World Wide Web Conference, pp. 2022–2032 (2019). [9] Yang, S., Song, G., Jin, Y. et al.: Domain adaptive classification on heterogeneous information networks. In: International Joint Conference on Artificial Intelligence, pp. 1410–1416 (2020). [10] Zhang, Y., Song, G., Du, L. et al.: Dane: Domain adaptive network embedding. In: International Joint Conference on Artificial Intelligence, pp. 4362–4368 (2019). [11] Ji, G., He, S., Xu, L. et al.: Knowledge graph embedding via dynamic mapping matrix. In: Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1, Long Papers) pp. 687–696 (2015). [12] Yang, B. Yih, W. He, X. et al.: Embedding entities and relations for learning and inference in knowledge bases. In: The 3rd on Learning Representations, (2015). International Conference [13] Trouillon, T., Welbl, J., Riedel, S. et al.: Complex embeddings for simple link prediction. In: International conference on machine learning, pp. 2071–2080 (2016). [14] Nickel, M., Tresp, V., Kriegel, H.-P. A three-way model for collective learning on multi-relational data. In: Proceedings of the 28th International Conference on International Conference on Machine Learning, pp. 809– 816 (2011). [15] Balazevic, I., Allen, C., Hospedales, T. Multi-relational in Neural poincare graph Information Processing Systems, 32, 4463– 4473 (2019). embeddings. Advances representation [16] Dong, Y., Chawla,N. V., Swami, A. Metapath2vec: Scalable for heterogeneous networks. In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 135–144 (2017). learning [17] Zhang, D., Yin, J., Zhu, X. et al.: Meta-graph2vec: Complex semantic path augmented heterogeneous network embedding. In: Pacific-Asia Conference on Knowledge Discovery and Data Mining. pp.196–208 (2018). [18] Kipf, T. N., Welling, M. Semi-supervised classification with graph convolutional networks. In: International Conference on Learning Representations (ICLR), (2017). [19] Wu, M., Pan, S., Zhou, C. et al.: Unsupervised domain adaptive graph convolutional networks. In: Proceedings of The Web Conference 2020, pp. 1457–1467 (2020). [20] Xu, K., Hu,W., Leskovec, J. et al.: How powerful are graph International Conference on In: neural networks?” Learning Representations, (2018). [21] Velicˇkovic´, P., Cucurull, G., Casanova, A. e t al . : Graph preprint networks. arXiv:1710.10903 (2017). attention arXiv [22] Huang, T., Xu, K., Wang, D. Da-hgt: Domain adaptive preprint transformer. arXiv heterogeneous arXiv:2012.05688 (2020). graph [23] Tzeng, E., Hoffman, J., Saenko, K. et al.: Adversarial discriminative domain adaptation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 7167–7176 (2017). [24] Tolstikhin, I.O., Sriperumbudur, B. K., Scho¨lkopf, B. Minimax estimation of maximum mean discrepancy with radial kernels. Advances in Neural Information Processing Systems, 29 pp. 1930–1938 (2016). [25] Fuglede, B.,Topsoe, F. Jensen-shannon divergence and hilbert space embedding. In: International Symposium on Information Theory, ISIT 2004, Proceedings. p. 31 (2004) . [26] Ding, Z., Li, S., Shao, M. et al.: Graph adaptive knowledge transfer for unsupervised domain adaptation. In: Proceedings of the European Conference on Computer Vision (ECCV), pp. 37–52 (2018). [27] Connor, R., Cardillo, F. A., Moss, R. et al.: Evaluation of Jensen-shannon distance over sparse data. In: International Conference on Similarity Search and Applications, pp. 163–168 (2013). [28] Ganin, Y., Ustinova, E., Ajakan, H. et al.: Domain- adversarial training of neural networks. The Journal of Machine Learning Research, 17(1), 2096–2030 (2016). [29] Goodfellow, I., Pouget-Abadie, J., Mirza, M. Generative adversarial networks. Communications of the ACM, 63(11), 139–144 (2020). [30] Hao, Y., Cao, X., Fang, Y. et al.: Inductive link prediction for nodes having only attribute information. arXiv preprint arXiv:2007.08053 (2020). [31] Dziugaite, G.K., Roy, D.M. Computing nonvacuous generalization bounds for deep (stochastic) neural networks with many more parameters than training data. arXiv preprint arXiv:1703.11008 (2017). [32] Han, X., Cao, S., Lv, X. et al.: Openke: An open toolkit for knowledge embedding. In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pp. 139–144 (2018). [33] Wang, M., Yu, L., Zheng, D. et al.: Deep graph library: Towards efficient and scalable deep learning on graphs. In: ICLR Workshop on Representation Learning on Graphs and Manifolds. 2019. APPENDIX A. Relation Decomposition of WN18 The sources A, B, and C of WN18 are decomposed by relations according to their semantic meaning, where the names of the relations for each source can be found below: • A: instance hyponym, hyponym, hypernym, member holonym, instance hypernym, member meronym l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i / . t / 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d . / t i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 • B: member of domain topic, synset domain usage of, synset domain region of, member of domain region, derivationally related form, member of domain usage, synset domain topic of • C: part of, verb group, similar to, also see, has part B. Model Configurations We use adagrad as the optimizer with a learning rate of 0.005 and a weight decay of 0.001 for all models. Each positive edge is trained with four negative edges to compute the margin- based loss. The size of the minibatch is 1024. The embeddings of each experiment are trained with 2000 epochs and all the methods converged at this level. For link prediction, we train a matcher model for each experiment to be an MLP with two hidden layers of hidden dimension 200. The matcher model takes the concatenated head, relation and tail embeddings as the input and has output the softmax probability of having a link. For GNN matcher models (GCN/GAT/GIN), the number of layers is set to be 2 with the final dropout ratio to be 0.4. We train each of the matcher models for 200 epochs in each experiment. For node classification, we train an MLP classifier with 1 hidden layer of hidden dimension 200 and a softmax output layer for the probability of the six classes. We train the classifier in each experiment for 200 epochs. C. Implementation Details We implement our methods in Python. We utilize OpenKE [32] as the backend for loading triples to training and perform- ing link prediction evaluation using the trained embeddings. We also use the dgl library [33] to perform graph-related computations and PyTorch to perform neural network com- putations. The models are trained on a server equipped with four NVIDIA TESLA V100 GPUs. The codes and data for the paper are available and will be made public after this paper is published. D. Metrics • Link prediction metrics: – Mean reciprocal rank (MRR): Mean of reciprocal ranks of first relevant edge. Given a series of query testing edges 𝑄, and rank7 be the rank of a true edge over 1000 negative entities chosen, the MRR is computed by MRR = 1 |𝑄| |l| L 79: 1 rank7 – Mean rank (MR): Mean rank of the first relevant edge, subject to larger variance as the high-rank edges which contain diverged features dominate the mean of ranks. MR is computed by MR = 1 |𝑄| |l| L rank7 79: – Hit rate @ n: the fraction of positives that rank in the top n rankings among their negative samples. • Classification metrics: – Accuracy: The fraction of correct predictions to the total number of ground truth labels. E. Additional Visualizations Figure 5 presents the visualization results of the embeddings of entities from WN18 dataset, with and without distribution alignment, respectively. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . e d u d n / i t / l a r t i c e - p d f / d o i / i . / / t 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d t / . i f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Data Intelligence Just Accepted MS. https://doi.org/10.1162/dint_a_00200 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . (a) Source A No DA (b) Source B No DA (c) Source C No DA e d u d n / i t / l a r t i c e - p d f / d o i / i t / / . 1 0 1 1 6 2 d n _ a _ 0 0 2 0 0 2 0 7 0 9 2 3 d n _ a _ 0 0 2 0 0 p d t . / i (d) Source A With DA (e) Source B With DA (f) Source C With DA Fig. 5: Isomap plots of the embeddings of WN18 by sources A, B, and C, with and without distribution alignment (DA) respectively. The alignment method used is adversarial regularization. f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3

Télécharger le PDF