=Paper=
{{Paper
|id=Vol-3742/paper14
|storemode=property
|title=Course contrastive recommendation algorithm based on hypergraph convolution
|pdfUrl=https://ceur-ws.org/Vol-3742/paper14.pdf
|volume=Vol-3742
|authors=Yanlie Zheng,Xueying Li,Qingxia Shen
|dblpUrl=https://dblp.org/rec/conf/citi2/ZhengLS24
}}
==Course contrastive recommendation algorithm based on hypergraph convolution==
https://ceur-ws.org/Vol-3742/paper14.pdf
Course contrastive recommendation algorithm based on
hypergraph convolution
Yanlie Zheng 1,∗,†, Xueying Li 1,† and Qingxia Shen 1,†
1 Fiberhome Telecommunication Technologies Co.,LTD, Wuhan 430068, China
Abstract
With its powerful modeling ability of real data, graph-based convolutional recommendation
algorithms have become an important tool for providing personalized services to users.
However, the existing graph-based recommendation algorithms mainly face the following
problems: 1. Simple graph-based neural network models are difficult to this complex
interaction relationship between user-items and their interaction high order.2. Personalized
recommendation systems rely on users to generate their own data, which makes it difficult to
obtain effective labeling information.3. Interaction noise. User-item interaction is saturated
with noise interference. In order to overcome these problems and improve the performance of
personalized recommendation, in this paper, we design a course comparison recommendation
method based on enhanced hypergraph convolution (DHSL-Cu). Specifically, first, we design the
introduction of a dual hypergraph convolutional network to capture the higher-order
relationships between users and items and the potential interaction information under
different interaction types. Second, we design a momentum-driven twin-based architecture to
optimize the target network using a momentum-based parameter update strategy. Finally, a
negative sample selection mechanism based on course learning is designed as our training
strategy. Finally, through extensive experiments on real-world datasets and parameter analysis,
we demonstrate that our proposed DHSL-Cu model can efficiently improve recommendation
services.
Keywords
Hypergraph, Contrastive Learning, Recommendation, Course Learning 1
1. Introduction
In recent years, due to the powerful learning capability demonstrated by graph
convolutional networks in the field of non-Euclidean data, many researchers have
introduced them into the recommendation domain [1,2]. They consider users and items in
the recommendation domain as nodes of an item, and the network relationships formed
between nodes as the interactions between users and items, and learn the embedded
CITI’2024: 2nd International Workshop on Computer Information Technologies in Industry 4.0, June 12–14, 2024,
Ternopil, Ukraine
∗ Corresponding author.
† These authors contributed equally.
zhengyanlie@ fiberhome.com (Y. Zheng); xli@ fiberhome.com (X. Li); qshen@ fiberhome.com (Q. Shen)
0000-0003-3412-1639 (Y. Zheng); 0009-0008-2733-3565 (X. Li); 0009-0000-1600-4717 (Q. Shen)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
information of users and items embedded in the graph structure through convolution [3].
Yin et al. proposed to demonstrate the potential of the graph CNN approach in Pinterest's
recommendation task [4]. GC-MC [5] and NGCF [5] have constructed bipartite graphs of
user-item interactions from user-item interaction data in order to learn user preferences
by utilizing the user-item graph structure. Compared to the above simple graph models in
recommendation, HyperGCN [6] combines hypergraph with graph neural network,
introduces the concept of hypergraph convolution [7], and propagates the information of
nodes with the same interactions through hyperedge aggregation, which is able to
efficiently capture the higher-order interaction information between nodes.
Although the above work has made some progress, it is still subject to the following 2
limitations: 1) the personalized recommendation system relies on users to generate their
own data, which makes it difficult to obtain effective labeling information [8]. 2) It is well
known that GCN-based recommendation models are susceptible to the noise of user-item
interactions [9].
For this reason, in this paper, in order to alleviate the shortcomings in existing
methods, we propose a course comparison recommendation method based on enhanced
hypergraph convolution (DHSL-Cu). Specifically, firstly, in order to capture the complex
interactions between users and programs, we introduce a dual hypergraph convolutional
model to serve as an encoder to capture the higher-order relationships between user
programs and the potential interaction information under different interaction types.
Second, we design a momentum-driven twin architecture to optimize the target network
using a momentum-based parameter update strategy. Next, we introduce the idea of
course learning and design a negative sample selection mechanism based on course
learning as our training strategy. Through extensive experiments on real-world datasets
as well as parameter analysis, the effectiveness and superiority of the DHSL-Cu model are
demonstrated.
2. Model
Figure 1: Overall model framework
Figure 1 illustrates our model for DHSL-Cu. As shown, our model consists of three main
core modules:
Hypergraph Convolution Module: it is used to construct two-view and user-item
hypergraph sets. In this module, we adopt graph augmentation to obtain new views of
user-item interactions. After that, for two views, we construct user hypergraph set u and
G
G G'
item hypergraph set i for online view and user hypergraph set u and item hypergraph
G'
set i for target view based on the interaction information of users and items. Then,
similar to the literature [10], the characterization of users and items is carried out based
on the convolution operation.
Twin network-based course comparison learning module:
1. Momentum-driven twin network based module: in this module both online network
and target network use the gated dual hypergraph based convolutional model as the
network structure. Meanwhile, we use the momentum update mechanism to optimize
the target network's representation learning module during the parameter training
process, which aims to utilize the characteristics of the momentum update mechanism
to encode historical information in the target network, thus guiding the online network
to learn to explore richer and more effective feature representations. At the same time,
the asymmetry of the comparison network is enhanced by this parameter updating
method thus alleviating the possible training collapse problem.
2. Negative sampling training strategy based on course learning, meanwhile, in order to
solve the problem of too much randomness of negative sampling in contrast learning,
we introduce the method of course learning. In the negative sampling process first
through the Scoring function S ( ) , we sort the negative samples from easy to hard. In
addition, a Pacing function P ( ) is used to control the negative samples into the
training process. Then, the mutual and consistency information between the two views
is maximized, thus enhancing the user-item representation learning. Finally the user-
item feature representation learned through the whole model is used as the final user-
item representation for subsequent prediction tasks.
2.1. Formalization
U , I , E
Let a multipartite bipartite graph have two different types of node sets (user
U I
set and item set ) consisting of edge sets
E E 1 E 2 E K , where E denotes
j
the jth type of edge. For example, Amazon data user behavior log can be represented as, a
multiple bipartite graph containing two kinds of nodes (users, items) and two kinds of
edges (clicks, queries).
Input: Firstly, the original data is used to perform the corresponding data division and
processing, according to the interaction behavior between users and items, for the
embedding initialization of user-item nodes, to obtain the user-item initial feature matrix
X XU , X I X N F X I M F
, where U , denote the feature representations of the
user set and the item set, respectively, and F is the feature dimensionality, N , M are the
X U and item eigenvalues
number of users and the number of items. The user eigenvalues
X I are used as inputs to the model in terms of hypergraph sets.
Output: By feeding the corresponding data and structures to our model, the model
learns the final user-item representation for the next prediction recommendation task.
2.2. Hypergraph Building Blocks
We base our approach on two different enhancement methods to enhance the topological
and attribute information of the graphs in order to obtain a new view of the user-item
interaction. For both views we construct the user hypergraph set GU for the online view
and the user hypergraph set GU' for the target view based on the user set U and the user
hypergraph set GI for the original view and the project hypergraph set GI' for the target
view based on the project set I . Among them, by combining the edge modification policy
[11] is used to construct the online view, and we use the GD policy [12] for constructing
the target view.
We transform both views resulting from the above graph enhancement operations into
two isomorphic hypergraph sets. It is shown below:
(3)
= GU gU ,base , gU ,1..., gU ,k , GI g I ,base , g I ,1..., g I ,k ,
where gU , j U , U , j
and g I , , where
I, j I, j U, j and I , j denote the
hyperedges in the hypergraphs gU , j and g I , j , respectively. Note that all hypergraphs in
GU share the same set of user nodes U while all hypergraphs in GI share the same set of
project nodes I . For each project node i I , a hyperedge U , j is introduced in the
j
hypergraph gU , j connecting u | u U ,(u, i ) E , i.e., all user nodes in the set of user
nodes U that are directly connected to the project node i through the interaction type
E j . Similarly, for a user node u U , a hyperedge I , j is introduced in g I , j that connects
i | i I ,(u, i) E i.e., all project nodes in the set of project nodes I that are directly
j
connected to the user node u through the interaction type E j . Note that two special
k k
hypergraphs gU ,base GU and g I ,base GI are defined as g U , U , j and g V , I , j ,
j 1 j 1
i.e., the hypergraph is the one consisting of all interaction types between user items.
2.3. Dual Hypergraph Convolution Module
Dual Hypergraph Convolution Module mainly utilizes two chi-sub hypergraphs GU and
GI in online view or target view to obtain the potential features of users and items
behind different interaction types by aggregating and propagating the feature
representations of users and items through the hyperedge aggregation and propagation of
corresponding hypergraphs constructed on the basis of different interaction types in two
hypergraphs through convolution operation. The specific process is as follows:
First, we introduce an association matrix H in the hypergraph to describe the
relationship between nodes and hyperedges, given the hypergraph gU , j U , U , j ,
where j base,1,..., k denotes the different interaction types between user-items, and
k is the number of interaction types between user-items, and defines the association
matrix of gU , j as:
1 u e e U , j (4)
HU, j u, e
0 otherwise
U U , j
where H
U,j , U , j denotes the set of hyperedges in the hypergraph g U, j
,
where j base,1,..., k , By the same definition, we define the cross association matrix
|U , j ||U , j |
for the hypergraph g I , j . Let the diagonal matrices D u, u |U ||U | and B
U,j U,j
denote the node degree matrix and the hyperedge degree matrix, respectively. Where
D u,u
U, j H u, e and B e, e
e U , j U,j H u , e . We utilize the hypergraph
U,j uU U,j
spectral convolution operator to learn the embedding of each hypergraph in the model,
and the hypergraph convolution operator is denoted as:
X l 1 ( HWH T X l P l ) (5)
Then, we use the normalized hypergraph convolution operator and define the
hypergraph convolution operator for gU , j as:
X Ul ,1j ( DU1, j HU , jWU BU1, j HUT , j XUl , j PUl , j ) (6)
where denotes the nonlinear activation function, X l |U | dFl denotes the user
U,j
Fl Fl 1
features in layer l , WU |I ||I |
is a unitary matrix, PUl , j is a learnable
transformation matrix, and Fl and Fl 1 denote the embedding dimensions in layers l and
l 1 . For the dual isomorphic hypergraph sets GU and GI , we independently learn the
node features from each isomorphic hypergraph gU , j and g I , j . As a result, we obtain
node representations for users and projects based on different interaction types:
X U ,base , X U ,1 ,..., X U ,k and X I ,base , X I ,1 ,..., X I ,k .
Finally based on each layer of hypergraph convolution operator, after t rounds of
iterations, we can get the embeddings of users and items under different types of
interaction data, defined as X Ut X Ut ,base , X Ut ,1 , , X Ut ,k , X Ut , j R u Ft ,
X It X It ,base , X It ,1 ,, X It ,k , X It , j R i Ft . where Ft is the final embedding dimension
and k is the number of edge types. Finally, the feature values learned from multiple
interaction types are concatenated together to obtain the final embedding result as follows:
X U X Ut WU bU , X I X It WI bI (7)
k 1*Ft Ft Ft U Ft
where WU ,WI and bU , bI are trainable parameters. XU and
I Ft
XI . This allows us to obtain the final embedding Z X U , X I of users and items in
the online graph.
2.4. Momentum-driven course comparison based module
2.4.1. Momentum-driven twinning-based network structure
In this module we construct a momentum-driven twin network-based architecture as our
backbone, consisting of two identical network structures - the online network and the
target network - which use the same encoder - the hypergraph convolutional module.
Through the momentum optimization-based twin network feature, while the online
network encodes the node features, the historical training information is retained in the
target network, and then the mutual information between the two view representations is
maximized in the course-based comparative learning module, which guides the online
network to learn to explore richer and more effective feature representations.
The details are as follows: the target network uses the momentum update mechanism
in the way of updating the parameters. The purpose is to make the target graph retain part
of the historical information in the optimization process through the characteristics of the
momentum update mechanism, and guide the online graph to learn richer and more
effective feature representations in the comparison learning session. The iterative process
of its model parameters is as follows:
t aA
A t 1 1 a At (8)
where a is an adjustable parameter controlling the degree of temporary information
t denote the learnable parameters of the encoder at round t for
retention, and At and A
the online network and target network, respectively. Thus, the embedding of users and
items in the target graph is obtained Z X U , X I .
2.4.2. Negative Sampling Module Based on Course Learning
Aiming at the problems arising from the negative sampling strategies of existing
comparison learning papers, we design a novel negative sampling method based on course
learning. The main idea is to sort the negative samples according to the difficulty during
the training period, and introduce different negative samples into the training at different
stages according to the difference in the difficulty of the negative samples.
Similar to the literature [39], firstly, for the embedding result of any node in the online
graph, c negative embeddings {zc } with different difficulties can be found in the target
graph. meanwhile, a scoring function S is defined to map the different negative
embeddings into the target graph. that maps different negative embeddings to a numerical
score S zc to measure this difficulty. Further, the scoring function is set to sim zv , z c to
fully measure the difficulty of negative samples. The formula is as follows:
(9)
|| zz || zz ||
S z c
v
v c
c
S ( zc ) zv zc (10)
S zc sim zv , zc ( zv zc )T .1 ( zv zc ) (11)
where is the covariance matrix of multidimensional random variables.
In line with the literature [12], after we obtain the fraction S zc of each individual
negative embedding zc in the memory bank, we use the pacing function to schedule how
to introduce negative samples into the training process. The pacing function P(t )
specifies the size of the memory bank to be used at each step t . The memory bank for t
consists of the P(t ) lowest scoring samples. Negative sample batches are sampled
uniformly from this set. We denote the complete memory bank size by C and the total
number of training steps by T . The formula is as follows:
t (12)
P (t ) 1 .1log e 10 K
T
P(t ) t/T K
(13)
where is a smoothing parameter used to control the speed of the training process.
1/ 2,1, 2 denote the root, linear and quadratic pacing functions, respectively.
2.4.3. Contrastive loss
To maximize the consistency between the online view and the target view. We use noise to
estimate the contrast loss. Specifically, we define a "memory bank" Q , which contains
each positive pair zv , zv , i.e., the same node in both views, and C negative samples
C
embedded in {zc }c 1 , the different nodes in both views, and then we use the similarity
measure function sim , to compute the positive pair zv , zv and negative pair{ zv , z c }
, based on which the loss function is as follows:
exp sim zv , zv / (14)
lNCE log C
exp( sim( zv , zv ) / ) c 1 exp( sim( zv , zc ) / )
where denotes the temperature parameter. To simplify the calculation, we use dot
product as the similarity measure function.
2.5. Model Optimization
We first use the edges that already exist in the interaction data as positive edges and
extract some non-existing edges as negative edges. Finally, the loss function is designed by
maximizing the positive edge probability and minimizing the negative edge probability.
The loss function is defined as:
n
( u ,i )E j 1
lr log (ZuT Zi ) (1 ) E u j P (u ) log(1 ( ZuT Zu j )) E i j P (i ) log(1 (ZiT Zi j )) (15)
where is the sigmoid activation function, is the weight parameter in order to
balance the importance of positive and negative samples, P(u) defines the distribution of
u candidate nodes, and n is the number of negative samples, the existing edges in the
multipartite bipartite graph are used as positive samples, and for each positive sample of
edge (u, i) , n negative edges are randomly sampled from node u and node i .
Ultimately, we unify the representation learning module for the main recommendation
task and the self-supervised comparison learning module for auxiliary enhancement into
an overall learning framework. Formally, the final learning objective is defined as:
l lr l NCE (16)
where is a variable factor that controls the self-supervised contrast learning task.
Finally we train our model using the Adam algorithm.
3. Experimental Comparison and Analysis
3.1. Datasets
To evaluate the performance of our model, we conducted experiments on two real-world
datasets, Amazon [12] and Alibaba[12]. The relevant attribute statistics of the two
datasets are shown in Table 1.
Table 1
Statistics of the data set
Datasets User Item Interaction Interaction Density
types
Amazon 3781 5749 60658 2 0.279%
Alibaba 1869 13349 27036 3 0.108%
3.2. Experimental Setup
Consistent with the literature [13], we use AUROC, AUPRC, Precision and Recall as
evaluation metrics. Meanwhile, we randomly select 60% of the edges as the training set
and the rest of the edges as the test set. The whole process is repeated five times to obtain
different random samples for the training and test sets. The mean and standard deviation
values of the classification evaluation metrics are reported.
3.3. Experimental Design
In order to comprehensively test the performance of the DHSL-Cu model proposed in this
paper, we test it from different perspectives as follows.
1. comparison with mainstream advanced algorithms: in order to assess the effectiveness
of the DHSL-Cu model, this paper compares it with eight mainstream advanced
algorithms.
2. ablation analysis: the contribution of each component is analyzed.
3. parameter sensitivity analysis: the influence degree of momentum-driven update
weight ratio, learning rate, and self-supervised comparison learning factor is analyzed.
3.4. Baseline Algorithms
To evaluate the performance of the DHSL-Cu model, we compare it with the following
existing mainstream state-of-the-art recommendation algorithms.
1. GraphSAGE [13]: proposes a generalized induction architecture that uses local
neighbor sampling of nodes and aggregated features to generate embeddings of nodes.
2. GCN [13] uses spectral graph convolution operators to learn local graph network
structures and node features in order to achieve semi-supervised learning directly on
graph structure data.
3. GAT [13] uses masked self-attention to assign different weights to each node and its
neighboring nodes based on their features, eliminating the need to use a pre-
constructed graph.
4. HGNN [13] designed the hyperedge convolution operation to deal with data correlation
during representation learning. By this method, the hyperedge convolution operation
can be effectively utilized to capture the implicit layer representation of higher order
data structures.
5. HyperGCN [13] is a new method of GCN training for semi-supervised learning of
hypergraphs based on hypergraph theory.
6. DualHGCN [13] a self-supervised Dual Hypergraph Convolutional Network (DualHGCN)
model that transforms a multilayer two-part graph network into two sets of its
hypergraph sets.
7. SGL [13] designed three types of data augmentation based on different perspectives to
complement/supervise the recommendation task with self-supervised signals on user-
item graphs.
8. HCCF [13] designed hypergraph structure learning module and cross view hypergraph
contrast coding model based on contrast learning to learn better user representations
by characterizing both local and global collaborative relationships in joint embedding
space.
3.5. Analysis of experimental results
3.5.1. Comparison with baseline algorithm
Table 2
Comparison of overall performance in auroc, auprc, precision and recall
Methods Amazon Alibaba
auroc auprc precision recall auroc auprc precision recall
GraphSAGE 66.99 69.39 63.47 52.74 66.49 60.36 63.47 52.74
GCN 64.93 77.45 69.46 71.53 56.87 77.66 69.46 71.53
GAT 66.7 70.16 63.34 51.39 55.38 54.49 63.34 51.39
HGNN 80.14 82.94 78.54 69.51 69.64 73.5 78.54 69.51
HyperGCN 68.42 73.78 67.12 61.61 61.38 65.21 67.12 61.61
DualHGCN 83.46 88.69 85.63 76.39 84.57 86.02 85.63 76.39
HCCF 94.27 95.32 91.56 91.67 90.13 89.32 90.27 82.13
DHSL-Cu 96.68 97.89 93.7 94.68 93.57 91.76 91.76 84.94
The results of all algorithm experiments on both datasets are shown in Table 2. From the
experimental results of the four evaluation metrics AUROC, AUPRC, Precision and Recall,
we can conclude that:
1. GraphSAGE, GCN and GAT perform poorly on the two datasets, which may be due
to the fact that these embedding methods based on simple homogeneous graphs
have a weak interaction representation and do not deal well with non-planar
relationships between nodes compared to hypergraph convolution.
2. DualHGCN outperforms HGNN and HyperGCN, probably because HyperGCN and
HGNN are both based on the same type of interaction information, and when
confronted with complex user-item interactions under different types, it is difficult
to capture the potential higher-order information of the user and the item based on
different types of interactions. While DualHGCN effectively captures the interaction
information between different interaction types by designing the information
transfer mechanism between hypergraphs, the overall performance of DualHGCN
is weaker than that of HCCF, probably because HCCF is designed with a
hypergraph-based comparison learning model, which efficiently utilizes the local-
to-global cross-view supervision information.
In conclusion, DHSL-Cu outperforms the other benchmarks. This may be related to the
following two main reasons: 1) DHSL-Cu can effectively incorporate historical training
information by designing a momentum-driven target network structure, and the
parameter updating method of momentum update with gradient vanishing can effectively
alleviate the training collapse problem during the self-supervised learning process. 2)
During the training process, we use negative sampling based on course learning, which is
different from the general random sampling method, and it can effectively utilize local-to-
global cross-view supervision information. The negative sampling method based on
course learning is different from the general random sampling method, which can
effectively introduce different training phases according to different sample
characteristics, and effectively improve the generalization ability and prediction accuracy
of the model.
3.5.2. Ablation Analysis
In order to verify the different effects on the algorithm brought by the gating-based dual
hypergraph convolution module, the momentum-driven twin network structure, and the
negative sampling strategy based on course learning, we conducted experiments and
comparative analysis. The experimental results are shown in Table. 3, where DHSL-Cu-Init
denotes the feature initialization module only, DHSL-Cu-H introduces the gating-based
dual hypergraph convolution module after the feature priming module, and DHSL-Cu-M
introduces the momentum-driven twin-network-based module and comparative learning
on top of DHSL-Cu-H. DHSL-Cu-Both, i.e., the introduction of negative sampling strategy
after the introduction of the complete DHSL-Cu model.
Table 3
Comparative analysis of algorithmic impact
auroc auprc precision recall
amazon alibaba amazon alibaba amazon alibaba amazon alibaba
Init 59.50 72.37 64.77 59.00 63.48 48.73 57.36% 62.75
H 83.46 87.59 89.95 82.20 85.63 69.81 82.30 79.92
M 92.99 89.33 94.54 86.47 90.69 86.47 91.07 81.31
Both 96.68 93.59 97.89 91.76 93.70 91.76 94.98 84.94
As shown in Table. 3, DHSL-Cu-Init performs the worst. DHSL-Cu-H better and
accurately shows the effectiveness and efficiency of the gating-based dual hypergraph
mechanism than it. The improvement in the performance of DHSL-Cu-M and DHSL-Cu-
Both illustrates the power of the self-supervised contrast learning framework. Among
them, the best performance of DHSL-Cu-Both shows that the introduction of negative
sampling strategy can effectively improve and enhance the learning performance of
contrast learning, and DHSL-Cu and its variants outperform the Alibaba dataset on the
Amazon dataset. The possible reason could be that the Amazon dataset is denser, which
helps to capture more effective data during various types of user-item interactions.
3.5.3. Parameter sensitivity analysis
This section focuses on the effect of the factor setting for self-supervised learning on
the model performance.
As shown in Tables. 4, 5, the model performance at contrast learning factor less than
0.5 is improves with increasing contrast learning factor and reaches the optimal point for
all four evaluation metrics on the beta 0.5 two datasets, after which the performance
decreases with increasing contrast learning factor. This may be a result of learning loss
pairs that are too high in the gradient conflict between the prediction and comparison
tasks during training.
Table 4
Results of different contrasting learning factors on AUROC and AUPRC metrics
beta AUROC AUPRC
Amazon Alibaba Amazon Alibaba
0.1 96.39 92.29 97.74 90.91
0.2 96.39 92.67 97.70 90.94
0.3 96.44 92.56 97.76 90.56
0.4 96.53 93.36 97.79 90.53
0.5 96.64 93.57 97.86 91.44
0.6 96.50 93.24 97.77 90.78
0.7 96.51 93.09 97.74 90.40
0.8 96.43 93.07 97.77 90.89
0.9 96.49 93.15 97.74 90.53
Table 5
Results of different comparison learning factors on Precision and Recall metrics
beta Precision Recall
Amazon Alibaba Amazon Alibaba
0.1 93.12 88.85 94.25 82.03
0.2 93.27 89.24 94.35 81.86
0.3 92.79 88.83 94.43 82.73
0.4 93.39 89.15 94.44 83.48
0.5 93.53 89.47 94.48 84.93
0.6 93.34 88.48 94.44 83.45
0.7 93.33 88.31 94.21 82.14
0.8 93.17 87.59 94.31 82.23
0.9 93.23 88.91 94.21 81.57
4. Conclusion
We propose a new self-supervised learning method DHSL-Cu. Specifically, we first
generate two hypergraph views based on a two-part graph network of users and items
after 2 different graph enhancement strategies. For the target network in the twin
network we use a momentum-based parameter update mechanism. The slow moving
target network is made to encode the online network history observations. A course
comparison framework based on a negative sample selection mechanism for course
learning is also designed. Finally, the effectiveness and superiority of the proposed DHSL-
Cu model is well demonstrated through extensive experiments on real-world datasets as
well as parameter analysis.
References
[1] Xie F, Zhang Y, Przystupa K, Kochan O. A Knowledge Graph Embedding Based Service
Recommendation Method for Service-Based System Development [J]. Electronics,
2023, 12(13): 2935.
[2] Xu X, Przystupa K, Kochan O. Social Recommendation Algorithm Based on Self-
Supervised Hypergraph Attention [J]. Electronics, 2023, 12(4): 906.
[3] Saito Y, Yaginuma S, Nishino Y, et al. Unbiased recommender learning from missing-
not-at-random implicit feedback[C]//Proceedings of the 13th International
Conference on Web Search and Data Mining. 2020: 501-509.
[4] Zhu D, Zhang Z, Cui P, et al. Robust graph convolutional networks against adversarial
attacks[C]//Proceedings of the 25th ACM SIGKDD international conference on
knowledge discovery & data mining. 2019: 1399-1407.
[5] Wang X, He X, Wang M, et al. Neural graph collaborative filtering[C]//Proceedings of
the 42nd international ACM SIGIR conference on Research and development in
Information Retrieval. 2019: 165-174.
[6] Yadati N, Nimishakavi M, Yadav P, et al. Hypergcn: A new method for training graph
convolutional networks on hypergraphs[C]//Advances in neural information
processing systems, 2019, 32.
[7] Jiang K, Zhang C, Wei B, Li Z, Kochan O. Fault diagnosis of RV reducer based on
denoising time–frequency attention neural network [J]. Expert Systems with
Applications, 2024, 238: 121762.
[8] Yanqiao Zhu, Yichen Xu, Feng Yu, et al., Graph Contrastive Learning with Adaptive
Augmentation[C]// Proceedings of the Web Conference 2021, pp.2069–2080.
[9] Feng Y, You H, Zhang Z, et al. Hypergraph neural networks[C]//Proceedings of the
AAAI conference on artificial intelligence, 2019, 33(01): 3558-3565.
[10] Bian Shuqing, Zhao Xin Wayne, Zhou Kun, et al., Contrastive Curriculum Learning for
Sequential User Behavior Modeling via Data Augmentation[C]// Proceedings of the
30th ACM International Conference on Information & Knowledge Management, 2021,
pp.3737–3746.
[11] Qin Xiuyuan, Yuan Huanhuan, Zhao Pengpeng, et al., Meta-optimized Contrastive
Learning for Sequential Recommendation[C]// Proceedings of the 46th International
ACM SIGIR Conference on Research and Development in Information Retrieval, 2023,
pp 89–98.
[12] Xue H, Yang L, Rajan V, et al. Multiplex bipartite network embedding using dual
hypergraph convolutional networks[C]//Proceedings of the Web Conference 2021.
2021: 1649-1660.
[13] Wang Shoujin, Hu Liang, Wang Yan, et al., Graph Learning based Recommender
Systems: A Review[C]// Proceedings of the Thirtieth International Joint Conference
on Artificial Intelligence, 2021,pp. 4644-4652.