基于图神经网络的兴趣活动推荐算法

doi:10.13229/j.cnki.jdxbgxb20200068

吉林大学学报(工学版) ›› 2021, Vol. 51 ›› Issue (1): 278-284.doi: 10.13229/j.cnki.jdxbgxb20200068

• 计算机科学与技术 • 上一篇

基于图神经网络的兴趣活动推荐算法

魏晓辉^1,²(),孙冰怡^1,²,崔佳旭^1,²

^1.吉林大学计算机科学与技术学院，长春 130012
^2.吉林大学符号计算与知识工程教育部重点实验室，长春 130012

收稿日期:2020-02-10 出版日期:2021-01-01 发布日期:2021-01-20
作者简介:魏晓辉（1972-），男，教授，博士生导师. 研究方向：分布式计算，近似计算. E-mail: weixh@jlu.edu.cn
基金资助:
国家重点研发计划专项项目(2016YFB0201503);国家自然科学基金项目(U19A2061)

Recommending activity to users via deep graph neural network

Xiao-hui WEI^1,²(),Bing-yi SUN^1,²,Jia-xu CUI^1,²

^1.College of Computer Science and Technology，Jilin University，Changchun 130012，China
^2.Key Laboratory of Symbol Computation and Knowledge Engineering of Ministry of Education，Jilin University，Changchun 130012，China

Received:2020-02-10 Online:2021-01-01 Published:2021-01-20

摘要/Abstract

摘要：

针对在基于事件的社交网络中，用户和其参加过的活动天然构成异质网络这一特点，提出了一个端到端的推荐算法，旨在使用异质网络的高阶连接性和非线性匹配关系，提高活动的推荐命中率。首先，通过图神经网络提取异质图的高阶连接信息对每个节点的影响，更新节点的嵌入式表示；然后，将用户-活动的嵌入式表示输入到多层感知机中，得到基于当前嵌入式表示的匹配概率，反向传播更新模型参数；重复此过程，获得最终的匹配概率。实验结果表明：本文算法训练稳定；相较于已有方法，命中率提高10%以上，归一化折损累计增益提高约10%；相较于不考虑异质图的高阶连接性的情况，命中率和归一化折损累计增益均有提高。

Abstract:

In event-based social network， users and activities construct heterogeneous network naturally. We propose an end-to-end recommendation model using high-order connectivity in heterogeneous network and non-linear matching relations to increase hit ratio. First， we extract the effect of high-order connectivity of each node， and get the embedding of nodes by graph neural network. Then， the embedding of user-activity pair is input of the multilayer perceptron to gain the current matching probability. Finally， by repeating this process the final matching probability is obtained. Experiments show that， the recommendation method is stable. Comparing with existing algorithms， our algorithm increases hit ratio more than 10%， and normalized discounted cumulative gain about 10%. Comparing with using multilayer perceptron only， both hit ratio and normalized discounted cumulative gain rise when considering high-order connectivity information.

Key words: computer software and theory, recommendation system, event-based social network, heterogeneous graph, graph neural network

中图分类号:

TP301.6

魏晓辉,孙冰怡,崔佳旭. 基于图神经网络的兴趣活动推荐算法[J]. 吉林大学学报(工学版), 2021, 51(1): 278-284.

Xiao-hui WEI,Bing-yi SUN,Jia-xu CUI. Recommending activity to users via deep graph neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(1): 278-284.

图/表 7

图1

图2

表1

表2

图3

表3

表4

参考文献 17

1	Liu X, He Q, Tian Y, et al. Event-based social networks: linking the online and offline social worlds[C]∥Proceedings of the 18th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Beijing, China, 2012: 1032-1040.
2	Qiao Z, Zhang P, Cao Y, et al. Combining heterogenous social and geographical information for event recommendation[C]∥Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence, Quebec, Canada, 2014: 145-151.
3	Macedo A Q, Marinho L B, Santos R L T. Context-aware event recommendation in event-based social networks[C]∥Proceedings of the 9th ACM Conference on Recommender Systems, Vienna, Austria, 2015: 123-130.
4	Pham T, Li X, Cong G, et al. A general graph-based model for recommendation in event-based social networks[C]∥IEEE 31st International Conference on Data Engineering, Seoul, South Korea, 2015: 567-578.
5	Wu X, Dong Y, Shi B, et al. Who will attend this event together? event attendance prediction via deep LSTM networks[C]∥ Proceedings of the 2018 SIAM International Conference on Data Mining, San Diego, CA, USA, 2018: 180-188.
6	董立岩, 王越群, 贺嘉楠, 等. 基于时间衰减的协同过滤推荐算法[J]. 吉林大学学报:工学版, 2017, 47(4): 1268-1272.
	Dong Li-yan, Wang Yue-qun, He Jia-nan, et al. Collaborative filtering recommendation algorithm based on time decay[J]. Journal of Jilin University (Engineering and Technology Edition), 2017, 47(4): 1268-1272.
7	Hamilton W L, Ying R, Leskovec J. Representation learning on graphs: methods and applications[J]. arXiv:1709.05584, 2017: 1-24.
8	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, Montreal, QC, Canada, 2009: 452-461.
9	Wang H, Terrovitis M, Mamoulis N. Location recommendation in location-based social networks using user check-in data[C]∥Proceedings of the 21st ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, Orlando, Florida, USA, 2013: 364-373.
10	Jiang X, Sun X, Zhuge H. Towards an effective and unbiased ranking of scientific literature through mutual reinforcement[C]∥Proceedings of the 21st ACM International Conference on Information and Knowledge Management, Maui, HI, USA, 2012: 714-723.
11	Thomas N K, Max W. Semi-Supervised classification with graph convolutional networks[C]∥The 5th International Conference on Learning Representations, Toulon, France, 2017.
12	Hamilton W, Ying R, Leskovec J. Inductive representation learning on large graphs[C]∥Advances in Neural Information Processing Systems, Long Beach, CA, USA, 2017: 1024-1034.
13	Berg R V D, Kipf T, Welling M. Graph convolutional matrix completion[C]∥Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, London, UK, 2018.
14	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, Paris, France, 2019: 165-174.
15	He X, Liao L, Zhang H, et al. Neural collaborative filtering[C]∥Proceedings of the 26th International Conference on World Wide Web, Perth, Australia, 2017: 173-182.
16	Kingma D, Ba J. Adam: a method for stochastic optimization[C]∥The 3rd International Conference for Learning Representations, San Diego, USA, 2015.
17	Bayer I, He X, Kanagal B, et al. A generic coordinate descent framework for learning from implicit feedback[C]∥Proceedings of the 26th International Conference on World Wide Web, Perth, Australia, 2017: 1341-1350.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

城市	交互数	用户数	活动数	稀疏度
洛杉矶	112 073	7 588	10 273	0.001 54
休斯敦	10 469	774	1 465	0.009 91
纽约	443 199	29 059	31 003	0.000 50

参数名	参数值	参数名	参数值
迭代次数	35	批尺寸	512
邻居层数	3	学习率	0.001
嵌入式维度	64	惩罚率	0.0001
MLP维度	[128,64,32]

对比算法	洛杉矶		休斯敦		纽约
对比算法	HR @10	NDCG@10	HR @10	NDCG@10	HR @10	NDCG@10
BPR^[8]	0.1076	0.0478	0.1010	0.0502	0.00053	0.00017
HereRS^[4]	0.6042	0.5121	0.4592	0.4227	0.5225	0.4884
GC?MC^[13]	0.3735	0.2113	0.6029	0.3594	0.2583	0.1103
NGCN^[14]	0.6054	0.4111	0.7077	0.4364	0.3026	0.2009
DGNN?AR	0.8249	0.6062	0.7379	0.4442	0.7651	0.5472

情况	洛杉矶		休斯敦		纽约
情况	HR @10	NDCG@10	HR @10	NDCG@10	HR @10	NDCG@10
DGNN?AR?0	0.7457	0.5083	0.5046	0.2471	0.7242	0.4936
DGNN?AR?1	0.8008	0.5603	0.6370	0.3648	0.7452	0.5257
DGNN?AR?2	0.8184	0.5859	0.6999	0.4176	0.7579	0.5284
DGNN?AR?3	0.8249	0.6062	0.7379	0.4442	0.7651	0.5472
DGNN?AR?4	0.8029	0.5616	0.7365	0.4283	0.7536	0.5291

基于图神经网络的兴趣活动推荐算法

Recommending activity to users via deep graph neural network

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 17

相关文章 1

Metrics

本文评价

推荐阅读 0