使用图负采样的图卷积神经网络推荐算法

doi:10.19665/j.issn1001-2400.20230214

摘要/Abstract

摘要：

经过几年的快速发展,基于图卷积神经网络的协同过滤算法已经在许多推荐系统场景中取得了最好的表现。但是这些算法在采集负样本时大多仅仅采用简单的随机负采样方法,没有充分利用图结构信息。针对这一问题,提出了一种使用图负采样的图卷积神经网络推荐算法GCN-GNS。该算法首先构造用户-物品二部图,并利用图卷积神经网络获取节点嵌入向量;接下来通过基于深度优先搜索的随机游走方法获取同时包含近距离邻居物品节点和远方物品节点的游走物品节点序列;然后设计注意力层自适应学习游走序列中不同节点的权重,并按权重组成一个动态更新的虚拟负样本;最终利用该虚拟负样本对模型进行更高效的训练。实验结果显示,与对比算法相比,多数情况下GCN-GNS在三个真实公开数据集上都有更好的表现;这表明所提出的新的图负采样方法能够帮助GCN-GNS算法更充分地利用图结构信息,并最终提升物品推荐的效果。

关键词: 推荐系统, 协同过滤, 卷积神经网络, 图负采样

Abstract:

After several years of rapid development,the collaborative filtering algorithms based on graph convolutional neural networks have achieved the most advanced performance in many recommender system scenarios.However,most of these algorithms only use simple random negative sampling method when collecting negative samples,and do not make full use of graph structure information.To solve this problem,a graph convolution neural network for recommendation using graph negative sampling(GCN-GNS) is proposed.The algorithm first constructs a user-item bipartite graph and uses a graph convolution neural network to obtain the node embedding vector.Next,the depth-first random walk method is used to obtain the sequence of the wandering item nodes that includes both the neighboring item nodes and the distant item nodes.Then the attention layer is designed to learn the weights of different nodes in the walk sequence adaptively and a dynamically updated virtual negative sample is formed according to the weights.Finally,the virtual negative sample is used to train the model more efficiently.Experimental results show that the GCN-GNS performs better than other algorithms for comparison on three real public datasets in most cases,which indicates that the proposed novel graph negative sampling method can help the GCN-GNS model to make better use of the graph structure information,and ultimately improves the effect of item recommendation.

Key words: recommender systems, collaborative filtering, convolutional neural networks, graph negative sampling

中图分类号:

TP181

黄河源, 慕彩红, 方云飞, 刘逸. 使用图负采样的图卷积神经网络推荐算法[J]. 西安电子科技大学学报, 2024, 51(1): 86-99.

HUANG Heyuan, MU Caihong, FANG Yunfei, LIU Yi. Graph convolution neural network for recommendation using graph negative sampling[J]. Journal of Xidian University, 2024, 51(1): 86-99.

图/表 9

图1

图2

图3

表1

图4

图5

图6

图7

图8

参考文献 31

[1]	LAVANYA R, SINGH U, TYAGI V. A Comprehensive Survey on Movie Recommendation Systems[C]//2021 International Conference on Artificial Intelligence and Smart Systems(ICAIS). Piscataway:IEEE, 2021:532-536.
[2]	CHEN R, HUA Q Y, CHANG Y S, et al. A Survey of Collaborative Filtering-Based Recommender Systems:from Traditional Methods to Hybrid Methods Based on Social Networks[J]. IEEE Access, 2018, 6:64301-64320. doi: 10.1109/ACCESS.2018.2877208
[3]	KOREN Y, BELL R, VOLINSKY C. Matrix Factorization Techniques for Recommender Systems[J]. Computer, 2009, 42(8):30-37.
[4]	史加荣, 李金红. 新型深度矩阵分解及其在推荐系统中的应用[J]. 西安电子科技大学学报, 2022, 49(3):171-182.
	SHI Jiarong, LI Jinhong. Novel Deep Matrix Factorization and Its Application in the Recommendation System[J]. Journal of Xidian University, 2022, 49(3):171-182.
[5]	韩立锋, 陈莉, 史晓龙. 融合项目属性偏好的矩阵分解推荐模型[J]. 西安电子科技大学学报, 2022, 49(3):147-159.
	HAN Lifeng, CHEN Li, SHI Xiaolong. Matrix Decomposition Recommendation Model Incorporating Item Attribute Preference[J]. Journal of Xidian University, 2022, 49(3):147-159.
[6]	WU S W, SUN F, ZHANG W T, et al. Graph Neural Networks in Recommender Systems:a Survey[J]. ACM Computing Surveys(CSUR), 2023, 55(5):1-37.
[7]	KIPF T N, WELLING M. Semi-Supervised Classification with Graph Convolutional Networks[C]//5th International Conference on Learning Representations(ICLR 2017). La Jolla: ICLR, 2017:1-14.
[8]	VELICKOVIC P, CUCURULL G, CASANOVA A, et al. Graph Attention Networks[C]//6th International Conference on Learning Representations(ICLR 2018). La Jolla: ICLR, 2018:1-12.
[9]	YU J, YIN H, XIA X, et al.(2022). Are Graph Augmentations Necessary? Simple Graph Contrastive Learning for Recommendation[C]// Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval(SIGIR 2022). New York: ACM, 2022:1294-1303.
[10]	MIKOLOV T, SUTSKEVER I, CHEN K, et al. Distributed Representations of Words and Phrases and Their Compositionality[C]//Twenty-seventh Conference on Neural Information Processing Systems(NeurIPS 2013). San Diego: NIPS, 2013, 2:3111-3119.
[11]	PARK D H, CHANG Y. Adversarial Sampling and Training for Semi-Supervised Information Retrieval[C]// Proceedings of the World Wide Web Conference(WWW’19). New York: ACM, 2019:1443-1453.
[12]	WANG X, XU Y K, HE X N, et al.Reinforced Negative Sampling over Knowledge Graph for Recommendation[C]//Proceedings of the World Wide Web Conference(WWW’20). T New York: ACM, 2020:99-109.
[13]	YANG Z, DING M, XU Z, et al. Region or Global? A Principle for Negative Sampling in Graph-Based Recommendation[J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(6):6264-6277.
[14]	ZHOU Y, ZHENG H X, HUANG X, et al. Graph Neural Networks:Taxonomy, Advances, and Trends[J]. ACM Transactions on Intelligent Systems and Technology, 2022, 13(1):1-54.
[15]	WANG Y, LIU Z W, FAN Z W, et al. DSKReG:Differentiable Sampling on Knowledge Graph for Recommendation with Relational GNN[C]// Proceedings of the 30th ACM International Conference on Information & Knowledge Management(CIKM ‘21). New York: ACM, 2021:3513-3517.
[16]	YANG L W, LIU Z W, DOU Y T, et al. ConsisRec:Enhancing GNN for Social Recommendation via Consistent Neighbor Aggregation[C]// Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval(SIGIR 2021). New York: ACM, 2021:2141-2145.
[17]	HAMILTON W L, YING R, LESKOVEC J. Inductive representation learning on large graphs[C]//Thirty-first Conference on Neural Information Processing Systems(NIPS 2017). San Diego: NIPS, 2017:1025-1035.
[18]	SUN J N, ZHANG Y X, MA C, et al. Multi-Graph Convolution Collaborative Filtering[C]//2019 IEEE International Conference on Data Mining(ICDM 2019). Piscataway:IEEE, 2019:1306-1311.
[19]	BERG R V D, KIPF T N, WELLING M. Graph Convolutional Matrix Completion[C]//24th ACM SIGKDD Conference on Knowledge Discovery and Data Mining Deep Learning Day(KDD’18 Deep Learning Day). New York: ACM, 2018:1-9.
[20]	WANG X, HE X N, WANG M, et al. Neural Graph Collaborative Filtering[C]// Proceedings of the 42th International ACM SIGIR Conference on Research and Development in Information Retrieval(SIGIR 2019). New York: ACM, 2019:165-174.
[21]	HE X N, DENG K, WANG X, et al. LightGCN:Simplifying and Powering Graph Convolution Network for Recommendation[C]// Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval(SIGIR 2020). New York: ACM, 2020:639-648.
[22]	RENDLE S, FREUDENTHALER C, GANTNER Z, et al. BPR:Bayesian Personalized Ranking from Implicit Feedback[C]//Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence(UAI’09). Montreal: AUAI Press, 2009:452-461.
[23]	PEROZZI B, AL-RFOU R, SKIENA S. DeepWalk:Online Learning of Social Representations[C]//In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining(KDD '14). New York: ACM, 2014:701-710.
[24]	YING R, HE R N, CHEN K F, et al. Graph Convolutional Neural Networks for Web-Scale Recommender Systems[C]//In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining(KDD'18). New York: ACM, 2018:974-983.
[25]	CHEN J W, WANG C, ZHOU S, et al. SamWalker:Social Recommendation with Informative Sampling Strategy[C]//In Proceedings of the World Wide Web Conference(WWW’19). New York: ACM, 2019:228-239.
[26]	WU L, YANG Y H, ZHANG K, et al. Joint Item Recommendation and Attribute Inference:An Adaptive Graph Convolutional Network Approach[C]//Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval(SIGIR 2020). New York: ACM, 2020:679-688.
[27]	MU C, HUANG H, LIU Y, et al. Graph Convolutional Neural Network based on the Combination of Multiple Heterogeneous Graphs[C]//2022 International Conference on Data Mining Workshops(ICDMW). Piscataway:IEEE, 2022:732-740.
[28]	GROVER A, LESKOVEC J. Node2vec: Scalable Feature Learning for Networks[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining(KDD’16). New York: ACM, 2016:855-864.
[29]	GLOROT X, BENGIO Y. Understanding the Difficulty of Training Deep Feedforward Neural Networks[C]//International Conference on Artificial Intelligence and Statistics(AISTATS 2010). New York: PMLR, 2010:249-256.
[30]	YANG J H, CHEN C M, WANG C J, et al. HOP-rec:High-Order Proximity for Implicit Recommendation[C]//In Proceedings of the 12th ACM Conference on Recommender Systems(RecSys'18). New York: ACM, 2018:140-144.
[31]	LIU Z W, MENG L, JIANG F, et al. Deoscillated Adaptive Graph Collaborative Filtering[C]// Topological,Algebraic and Geometric Learning Workshops 2022. New York: PMLR, 2022:248-257.

数据集方法	Lastfm		Douban		Flixster
数据集方法	NDCG	Recall	NDCG	Recall	NDCG	Recall
MF^[3]	0.186 87	0.243 87	0.084 48	0.128 49	0.109 57	0.177 65
GCN^[7]	0.168 17	0.227 48	0.085 12	0.128 71	0.081 11	0.143 45
GC-MC^[19]	0.155 66	0.212 96	0.074 96	0.112 52	0.065 22	0.127 30
NGCF^[20]	0.188 49	0.251 77	0.085 25	0.127 45	0.117 43	0.196 97
DGCF^[31]	[0.203 02]	[0.262 29]	[0.098 95]	[0.145 59]	0.131 42	0.219 57
LightGCN^[21]	0.213 40	0.275 38	0.102 74	0.149 62	[0.128 11]	[0.212 95]
GCN-GNS	0.221 27	0.285 5	0.127 28	0.170 96	0.130 58	0.218 20